add new status code for refund not ready

fix(deps): patch reachable pip security advisories (aiohttp, pyjwt, python-multipart, pydantic-settings)

migrations/versions/c6d7e8f9a0b1_add_slug_to_upstream_providers.py

@@ -0,0 +1,88 @@
+"""add slug to upstream_providers
+
+Revision ID: c6d7e8f9a0b1
+Revises: b5e7c9d1f3a2
+Create Date: 2026-06-29 00:00:00.000000
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+from routstr.core.provider_slugs import provider_slug_base, provider_slug_candidate
+
+revision = "c6d7e8f9a0b1"
+down_revision = "b5e7c9d1f3a2"
+branch_labels = None
+depends_on = None
+
+
+def _allocate_backfill_slug(provider_type: str, reserved_slugs: set[str]) -> str:
+    base = provider_slug_base(provider_type)
+    suffix_number = 1
+    while True:
+        candidate = provider_slug_candidate(base, suffix_number)
+        if candidate not in reserved_slugs:
+            reserved_slugs.add(candidate)
+            return candidate
+        suffix_number += 1
+
+
+def _backfill_provider_slugs(conn: sa.Connection) -> None:
+    existing_rows = conn.execute(
+        sa.text(
+            "SELECT slug FROM upstream_providers "
+            "WHERE slug IS NOT NULL AND slug != ''"
+        )
+    )
+    reserved_slugs = {str(row.slug).lower() for row in existing_rows}
+
+    rows_to_backfill = conn.execute(
+        sa.text(
+            "SELECT id, provider_type FROM upstream_providers "
+            "WHERE slug IS NULL OR slug = '' "
+            "ORDER BY id"
+        )
+    )
+    for row in rows_to_backfill:
+        slug = _allocate_backfill_slug(str(row.provider_type), reserved_slugs)
+        conn.execute(
+            sa.text("UPDATE upstream_providers SET slug = :slug WHERE id = :id"),
+            {"slug": slug, "id": row.id},
+        )
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    columns = {c["name"] for c in inspector.get_columns("upstream_providers")}
+
+    if "slug" not in columns:
+        op.add_column(
+            "upstream_providers",
+            sa.Column("slug", sa.String(), nullable=True),
+        )
+
+    _backfill_provider_slugs(conn)
+
+    existing_indexes = {idx["name"] for idx in inspector.get_indexes("upstream_providers")}
+    if "ix_upstream_providers_slug" not in existing_indexes:
+        op.create_index(
+            "ix_upstream_providers_slug",
+            "upstream_providers",
+            ["slug"],
+            unique=True,
+        )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    inspector = sa.inspect(conn)
+    existing_indexes = {idx["name"] for idx in inspector.get_indexes("upstream_providers")}
+    if "ix_upstream_providers_slug" in existing_indexes:
+        op.drop_index("ix_upstream_providers_slug", table_name="upstream_providers")
+
+    columns = {c["name"] for c in inspector.get_columns("upstream_providers")}
+    if "slug" in columns:
+        op.drop_column("upstream_providers", "slug")

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "routstr"
-version = "0.4.3"
+version = "0.4.4"
 description = "Payment proxy for your LLM endpoint using cashu and nostr."
 readme = "README.md"
 requires-python = ">=3.11"

routstr/auth.py

@@ -1021,32 +1021,37 @@ async def adjust_payment_for_tokens(
             # actual cost exceeded discounted reservation (due to tolerance_percentage)
             if cost_difference > 0:
                 # Always release the reservation and charge min(actual_cost, balance).
-                # Using a CASE expression makes this a single atomic UPDATE — no
-                # multi-level fallback needed and balance can never go negative.
+                # CASE expressions keep this atomic and safe even when the
+                # stale-reservation sweeper has already released the reservation.
                 chargeable = case(
                     (col(ApiKey.balance) >= total_cost_msats, total_cost_msats),
                     else_=col(ApiKey.balance),
                 )
+                overrun_safe_reserved = case(
+                    (
+                        col(ApiKey.reserved_balance) >= deducted_max_cost,
+                        col(ApiKey.reserved_balance) - deducted_max_cost,
+                    ),
+                    else_=0,
+                )
 
                 finalize_stmt = (
                     update(ApiKey)
                     .where(col(ApiKey.hashed_key) == billing_key.hashed_key)
-                    .where(col(ApiKey.reserved_balance) >= deducted_max_cost)
                     .values(
-                        reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost,
+                        reserved_balance=overrun_safe_reserved,
                         balance=col(ApiKey.balance) - chargeable,
                         total_spent=col(ApiKey.total_spent) + chargeable,
                     )
                 )
-                result = await session.exec(finalize_stmt)  # type: ignore[call-overload]
+                await session.exec(finalize_stmt)  # type: ignore[call-overload]
 
                 if billing_key.hashed_key != key.hashed_key:
                     child_stmt = (
                         update(ApiKey)
                         .where(col(ApiKey.hashed_key) == key.hashed_key)
-                        .where(col(ApiKey.reserved_balance) >= deducted_max_cost)
                         .values(
-                            reserved_balance=col(ApiKey.reserved_balance) - deducted_max_cost,
+                            reserved_balance=overrun_safe_reserved,
                             total_spent=col(ApiKey.total_spent) + min(billing_key.balance, total_cost_msats),
                         )
                     )
@@ -1054,51 +1059,38 @@ async def adjust_payment_for_tokens(
 
                 await session.commit()
 
-                if result.rowcount:
-                    await session.refresh(billing_key)
-                    if billing_key.hashed_key != key.hashed_key:
-                        await session.refresh(key)
-                    cost.total_msats = total_cost_msats
-                    logger.info(
-                        "Finalized payment with additional charge",
-                        extra={
-                            "key_hash": key.hashed_key[:8] + "...",
-                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
-                            "charged_amount": total_cost_msats,
-                            "new_balance": billing_key.balance,
-                            "model": model,
-                        },
-                    )
-                    await _accumulate_fee(total_cost_msats)
-                    payments_logger.info(
-                        "FINALIZE",
-                        extra={
-                            "event": "finalize",
-                            "key_hash": key.hashed_key[:8] + "...",
-                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
-                            "model": model,
-                            "cost_reserved": deducted_max_cost,
-                            "cost_charged": total_cost_msats,
-                            "input_tokens": cost.input_tokens,
-                            "output_tokens": cost.output_tokens,
-                            "balance": billing_key.balance,
-                            "reserved_balance": billing_key.reserved_balance,
-                            "total_spent": billing_key.total_spent,
-                            "finalize_type": "overrun",
-                        },
-                    )
-                else:
-                    # Guard fired: reservation was already released by a concurrent
-                    # finalization for this key. Nothing left to do.
-                    logger.warning(
-                        "Finalization skipped - reservation already released",
-                        extra={
-                            "key_hash": key.hashed_key[:8] + "...",
-                            "billing_key_hash": billing_key.hashed_key[:8] + "...",
-                            "attempted_charge": total_cost_msats,
-                            "model": model,
-                        },
-                    )
+                await session.refresh(billing_key)
+                if billing_key.hashed_key != key.hashed_key:
+                    await session.refresh(key)
+                cost.total_msats = total_cost_msats
+                logger.info(
+                    "Finalized payment with additional charge",
+                    extra={
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "charged_amount": total_cost_msats,
+                        "new_balance": billing_key.balance,
+                        "model": model,
+                    },
+                )
+                await _accumulate_fee(total_cost_msats)
+                payments_logger.info(
+                    "FINALIZE",
+                    extra={
+                        "event": "finalize",
+                        "key_hash": key.hashed_key[:8] + "...",
+                        "billing_key_hash": billing_key.hashed_key[:8] + "...",
+                        "model": model,
+                        "cost_reserved": deducted_max_cost,
+                        "cost_charged": total_cost_msats,
+                        "input_tokens": cost.input_tokens,
+                        "output_tokens": cost.output_tokens,
+                        "balance": billing_key.balance,
+                        "reserved_balance": billing_key.reserved_balance,
+                        "total_spent": billing_key.total_spent,
+                        "finalize_type": "overrun",
+                    },
+                )
             else:
                 # Refund some of the base cost
                 refund = abs(cost_difference)
@@ -1305,6 +1297,35 @@ async def periodic_key_reset() -> None:
             logger.error(f"Error in periodic_key_reset: {e}")
 
 
+async def periodic_dead_key_prune() -> None:
+    """Periodically prune dead API keys. Interval <= 0 disables it.
+
+    See ``prune_dead_api_keys`` for eligibility.
+    """
+    from .core.db import create_session, prune_dead_api_keys
+
+    interval = settings.dead_key_prune_interval_seconds
+    if interval <= 0:
+        logger.info("Dead-key pruning disabled (interval <= 0)")
+        return
+
+    while True:
+        try:
+            await asyncio.sleep(interval)
+        except asyncio.CancelledError:
+            break
+
+        try:
+            async with create_session() as session:
+                await prune_dead_api_keys(
+                    session, settings.dead_key_min_age_seconds
+                )
+        except asyncio.CancelledError:
+            break
+        except Exception as e:
+            logger.error(f"Error in periodic_dead_key_prune: {e}")
+
+
 STALE_RESERVATION_SWEEP_INTERVAL_SECONDS: int = 60
 
 

routstr/balance.py

@@ -259,11 +259,26 @@ async def refund_wallet_endpoint(
             )
         )
         in_tx = in_tx_result.first()
+        cashu_fingerprint = hashlib.sha256(x_cashu.encode()).hexdigest()[:16]
         if in_tx is None:
+            logger.info(
+                "refund_wallet_endpoint: x-cashu inbound transaction not found",
+                extra={"cashu_fingerprint": cashu_fingerprint},
+            )
             raise HTTPException(status_code=404, detail="Refund not found")
 
         # Use the request_id to find the associated "out" (refund) transaction
         if in_tx.request_id is None:
+            logger.info(
+                "refund_wallet_endpoint: x-cashu inbound transaction has no request_id",
+                extra={
+                    "cashu_fingerprint": cashu_fingerprint,
+                    "cashu_transaction_id": in_tx.id,
+                    "amount": in_tx.amount,
+                    "unit": in_tx.unit,
+                    "mint_url": in_tx.mint_url,
+                },
+            )
             raise HTTPException(status_code=404, detail="Refund not found")
 
         out_tx_result = await session.exec(
@@ -274,8 +289,33 @@ async def refund_wallet_endpoint(
         )
         out_tx = out_tx_result.first()
         if out_tx is None:
-            raise HTTPException(status_code=404, detail="Refund not found")
+            logger.info(
+                "refund_wallet_endpoint: x-cashu refund transaction not ready",
+                extra={
+                    "cashu_fingerprint": cashu_fingerprint,
+                    "cashu_transaction_id": in_tx.id,
+                    "refund_request_id": in_tx.request_id,
+                    "amount": in_tx.amount,
+                    "unit": in_tx.unit,
+                    "mint_url": in_tx.mint_url,
+                    "age_seconds": max(0, int(time.time()) - in_tx.created_at),
+                },
+            )
+            raise HTTPException(
+                status_code=425,
+                detail="Refund not ready. Retry later.",
+                headers={"Retry-After": "5"},
+            )
         if out_tx.swept:
+            logger.info(
+                "refund_wallet_endpoint: x-cashu refund transaction already swept",
+                extra={
+                    "cashu_fingerprint": cashu_fingerprint,
+                    "cashu_transaction_id": in_tx.id,
+                    "refund_transaction_id": out_tx.id,
+                    "refund_request_id": in_tx.request_id,
+                },
+            )
             raise HTTPException(status_code=410, detail="Refund has been swept")
 
         out_tx.collected = True

routstr/core/admin.py

@@ -1,5 +1,6 @@
 import asyncio
 import json
+import re
 import secrets
 from datetime import datetime, timezone
 from pathlib import Path
@@ -8,6 +9,7 @@ from fastapi import APIRouter, Depends, HTTPException, Query, Request
 from pydantic import BaseModel, RootModel
 from pydantic.v1 import ValidationError as PydanticValidationError
 from sqlmodel import select
+from sqlmodel.ext.asyncio.session import AsyncSession
 
 from ..payment.models import _row_to_model, list_models
 from ..proxy import refresh_model_maps, reinitialize_upstreams
@@ -29,6 +31,7 @@ from .db import (
 )
 from .log_manager import log_manager
 from .logging import get_logger
+from .provider_slugs import allocate_unique_provider_slug
 from .settings import SettingsService, settings
 
 logger = get_logger(__name__)
@@ -456,19 +459,18 @@ class ModelCreate(BaseModel):
     dependencies=[Depends(require_admin_api)],
 )
 async def upsert_provider_model(
-    provider_id: int, payload: ModelCreate
+    provider_id: str, payload: ModelCreate
 ) -> dict[str, object]:
     print(payload)
     logger.info(
         f"UPSERT_PROVIDER_MODEL called: provider_id={provider_id}, model_id={payload.id}"
     )
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        provider_pk = _provider_pk(provider)
 
         # Try to get existing model
-        existing_row = await session.get(ModelRow, (payload.id, provider_id))
+        existing_row = await session.get(ModelRow, (payload.id, provider_pk))
 
         if existing_row:
             # Update existing model
@@ -524,7 +526,7 @@ async def upsert_provider_model(
                 alias_ids=(
                     json.dumps(payload.alias_ids) if payload.alias_ids else None
                 ),
-                upstream_provider_id=provider_id,
+                upstream_provider_id=provider_pk,
                 enabled=payload.enabled,
                 forwarded_model_id=payload.forwarded_model_id or payload.id,
             )
@@ -543,7 +545,7 @@ async def upsert_provider_model(
     dependencies=[Depends(require_admin_api)],
 )
 async def update_provider_model_legacy(
-    provider_id: int, model_id: str, payload: ModelCreate
+    provider_id: str, model_id: str, payload: ModelCreate
 ) -> dict[str, object]:
     """Legacy PATCH endpoint - redirects to upsert POST endpoint for backward compatibility."""
     logger.info(
@@ -556,13 +558,12 @@ async def update_provider_model_legacy(
     "/api/upstream-providers/{provider_id}/models/{model_id:path}",
     dependencies=[Depends(require_admin_api)],
 )
-async def get_provider_model(provider_id: int, model_id: str) -> dict[str, object]:
+async def get_provider_model(provider_id: str, model_id: str) -> dict[str, object]:
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        provider_pk = _provider_pk(provider)
 
-        row = await session.get(ModelRow, (model_id, provider_id))
+        row = await session.get(ModelRow, (model_id, provider_pk))
         if not row:
             raise HTTPException(
                 status_code=404, detail="Model not found for this provider"
@@ -576,9 +577,11 @@ async def get_provider_model(provider_id: int, model_id: str) -> dict[str, objec
     "/api/upstream-providers/{provider_id}/models/{model_id:path}",
     dependencies=[Depends(require_admin_api)],
 )
-async def delete_provider_model(provider_id: int, model_id: str) -> dict[str, object]:
+async def delete_provider_model(provider_id: str, model_id: str) -> dict[str, object]:
     async with create_session() as session:
-        row = await session.get(ModelRow, (model_id, provider_id))
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        provider_pk = _provider_pk(provider)
+        row = await session.get(ModelRow, (model_id, provider_pk))
         if not row:
             raise HTTPException(
                 status_code=404, detail="Model not found for this provider"
@@ -593,10 +596,12 @@ async def delete_provider_model(provider_id: int, model_id: str) -> dict[str, ob
     "/api/upstream-providers/{provider_id}/models",
     dependencies=[Depends(require_admin_api)],
 )
-async def delete_all_provider_models(provider_id: int) -> dict[str, object]:
+async def delete_all_provider_models(provider_id: str) -> dict[str, object]:
     async with create_session() as session:
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        provider_pk = _provider_pk(provider)
         result = await session.exec(
-            select(ModelRow).where(ModelRow.upstream_provider_id == provider_id)
+            select(ModelRow).where(ModelRow.upstream_provider_id == provider_pk)
         )  # type: ignore
         rows = result.all()
         for row in rows:
@@ -615,7 +620,7 @@ class BatchOverrideRequest(BaseModel):
     dependencies=[Depends(require_admin_api)],
 )
 async def batch_override_provider_models(
-    provider_id: int, payload: BatchOverrideRequest
+    provider_id: str, payload: BatchOverrideRequest
 ) -> dict[str, object]:
     """Batch override models for a specific provider."""
     logger.info(
@@ -623,15 +628,14 @@ async def batch_override_provider_models(
     )
 
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        provider_pk = _provider_pk(provider)
 
         overridden_count = 0
 
         for model_data in payload.models:
             # Try to get existing model regardless of whether it's enabled or not
-            existing_row = await session.get(ModelRow, (model_data.id, provider_id))
+            existing_row = await session.get(ModelRow, (model_data.id, provider_pk))
 
             if existing_row:
                 # Update existing
@@ -685,7 +689,7 @@ async def batch_override_provider_models(
                         if model_data.alias_ids
                         else None
                     ),
-                    upstream_provider_id=provider_id,
+                    upstream_provider_id=provider_pk,
                     enabled=model_data.enabled,
                 )
                 session.add(row)
@@ -702,6 +706,85 @@ async def batch_override_provider_models(
     }
 
 
+_SLUG_PATTERN = re.compile(r"^[a-z0-9][a-z0-9-]{1,62}[a-z0-9]$")
+
+
+def _validate_slug(value: str) -> str:
+    candidate = value.strip().lower()
+    if not _SLUG_PATTERN.fullmatch(candidate):
+        raise HTTPException(
+            status_code=400,
+            detail=(
+                "slug must be 3-64 chars, lowercase letters/digits/hyphens, "
+                "and may not start or end with a hyphen"
+            ),
+        )
+    if candidate.isdigit():
+        raise HTTPException(
+            status_code=400,
+            detail="slug must not be all digits",
+        )
+    return candidate
+
+
+async def _ensure_unique_slug(
+    session: AsyncSession, slug: str, exclude_id: int | None = None
+) -> None:
+    stmt = select(UpstreamProviderRow).where(UpstreamProviderRow.slug == slug)
+    result = await session.exec(stmt)
+    existing = result.first()
+    if existing and existing.id != exclude_id:
+        raise HTTPException(
+            status_code=409,
+            detail="Provider with this slug already exists",
+        )
+
+
+async def _get_upstream_provider_by_ref(
+    session: AsyncSession, provider_ref: str
+) -> UpstreamProviderRow:
+    if provider_ref.isdigit():
+        provider = await session.get(UpstreamProviderRow, int(provider_ref))
+    else:
+        slug = _validate_slug(provider_ref)
+        result = await session.exec(
+            select(UpstreamProviderRow).where(UpstreamProviderRow.slug == slug)
+        )
+        provider = result.first()
+
+    if not provider:
+        raise HTTPException(status_code=404, detail="Provider not found")
+    return provider
+
+
+def _provider_pk(provider: UpstreamProviderRow) -> int:
+    if provider.id is None:
+        raise HTTPException(status_code=500, detail="Provider has no database id")
+    return provider.id
+
+
+def _serialize_provider(
+    provider: UpstreamProviderRow, redact_api_key: bool = True
+) -> dict[str, object]:
+    return {
+        "id": provider.id,
+        "slug": provider.slug,
+        "provider_type": provider.provider_type,
+        "base_url": provider.base_url,
+        "api_key": "[REDACTED]"
+        if (redact_api_key and provider.api_key)
+        else provider.api_key
+        if not redact_api_key
+        else "",
+        "api_version": provider.api_version,
+        "enabled": provider.enabled,
+        "provider_fee": provider.provider_fee,
+        "provider_settings": json.loads(provider.provider_settings)
+        if provider.provider_settings
+        else None,
+    }
+
+
 class UpstreamProviderCreate(BaseModel):
     provider_type: str
     base_url: str
@@ -710,6 +793,7 @@ class UpstreamProviderCreate(BaseModel):
     enabled: bool = True
     provider_fee: float = 1.01
     provider_settings: dict | None = None
+    slug: str | None = None
 
 
 class UpstreamProviderUpdate(BaseModel):
@@ -720,6 +804,50 @@ class UpstreamProviderUpdate(BaseModel):
     enabled: bool | None = None
     provider_fee: float | None = None
     provider_settings: dict | None = None
+    slug: str | None = None
+
+
+class UpstreamProviderUpdateBySlug(BaseModel):
+    slug: str
+    new_slug: str | None = None
+    provider_type: str | None = None
+    base_url: str | None = None
+    api_key: str | None = None
+    api_version: str | None = None
+    enabled: bool | None = None
+    provider_fee: float | None = None
+    provider_settings: dict | None = None
+
+
+async def _apply_provider_update(
+    session: AsyncSession,
+    provider: UpstreamProviderRow,
+    payload: UpstreamProviderUpdate,
+    new_slug: str | None = None,
+) -> None:
+    if new_slug is not None:
+        validated = _validate_slug(new_slug)
+        await _ensure_unique_slug(session, validated, exclude_id=provider.id)
+        provider.slug = validated
+
+    if payload.provider_type is not None:
+        provider.provider_type = payload.provider_type
+    if payload.base_url is not None:
+        provider.base_url = payload.base_url
+    if payload.api_key is not None:
+        provider.api_key = payload.api_key
+    if payload.api_version is not None:
+        provider.api_version = payload.api_version
+    if payload.enabled is not None:
+        provider.enabled = payload.enabled
+    if payload.provider_fee is not None:
+        provider.provider_fee = payload.provider_fee
+    if payload.provider_settings is not None:
+        provider.provider_settings = json.dumps(payload.provider_settings)
+
+    session.add(provider)
+    await session.commit()
+    await session.refresh(provider)
 
 
 @admin_router.get("/api/upstream-providers", dependencies=[Depends(require_admin_api)])
@@ -727,21 +855,7 @@ async def get_upstream_providers() -> list[dict[str, object]]:
     async with create_session() as session:
         result = await session.exec(select(UpstreamProviderRow))
         providers = result.all()
-        return [
-            {
-                "id": p.id,
-                "provider_type": p.provider_type,
-                "base_url": p.base_url,
-                "api_key": "[REDACTED]" if p.api_key else "",
-                "api_version": p.api_version,
-                "enabled": p.enabled,
-                "provider_fee": p.provider_fee,
-                "provider_settings": json.loads(p.provider_settings)
-                if p.provider_settings
-                else None,
-            }
-            for p in providers
-        ]
+        return [_serialize_provider(p) for p in providers]
 
 
 @admin_router.post("/api/upstream-providers", dependencies=[Depends(require_admin_api)])
@@ -761,7 +875,14 @@ async def create_upstream_provider(
                 detail="Provider with this base URL and API key already exists",
             )
 
+        if payload.slug:
+            slug = _validate_slug(payload.slug)
+            await _ensure_unique_slug(session, slug)
+        else:
+            slug = await allocate_unique_provider_slug(session, payload.provider_type)
+
         provider = UpstreamProviderRow(
+            slug=slug,
             provider_type=payload.provider_type,
             base_url=payload.base_url,
             api_key=payload.api_key,
@@ -778,99 +899,81 @@ async def create_upstream_provider(
 
     await reinitialize_upstreams()
     await refresh_model_maps()
-    return {
-        "id": provider.id,
-        "provider_type": provider.provider_type,
-        "base_url": provider.base_url,
-        "api_key": "[REDACTED]",
-        "api_version": provider.api_version,
-        "enabled": provider.enabled,
-        "provider_fee": provider.provider_fee,
-        "provider_settings": payload.provider_settings,
-    }
+    return _serialize_provider(provider)
 
 
 @admin_router.get(
     "/api/upstream-providers/{provider_id}", dependencies=[Depends(require_admin_api)]
 )
-async def get_upstream_provider(provider_id: int) -> dict[str, object]:
+async def get_upstream_provider(provider_id: str) -> dict[str, object]:
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
-        return {
-            "id": provider.id,
-            "provider_type": provider.provider_type,
-            "base_url": provider.base_url,
-            "api_key": "[REDACTED]" if provider.api_key else "",
-            "api_version": provider.api_version,
-            "enabled": provider.enabled,
-            "provider_fee": provider.provider_fee,
-            "provider_settings": json.loads(provider.provider_settings)
-            if provider.provider_settings
-            else None,
-        }
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        return _serialize_provider(provider)
 
 
 @admin_router.patch(
     "/api/upstream-providers/{provider_id}", dependencies=[Depends(require_admin_api)]
 )
 async def update_upstream_provider(
-    provider_id: int, payload: UpstreamProviderUpdate
+    provider_id: str, payload: UpstreamProviderUpdate
 ) -> dict[str, object]:
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
 
-        if payload.provider_type is not None:
-            provider.provider_type = payload.provider_type
-        if payload.base_url is not None:
-            provider.base_url = payload.base_url
-        if payload.api_key is not None:
-            provider.api_key = payload.api_key
-        if payload.api_version is not None:
-            provider.api_version = payload.api_version
-        if payload.enabled is not None:
-            provider.enabled = payload.enabled
-        if payload.provider_fee is not None:
-            provider.provider_fee = payload.provider_fee
-        if payload.provider_settings is not None:
-            provider.provider_settings = json.dumps(payload.provider_settings)
-
-        session.add(provider)
-        await session.commit()
-        await session.refresh(provider)
+        await _apply_provider_update(session, provider, payload, new_slug=payload.slug)
 
     await reinitialize_upstreams()
     await refresh_model_maps()
-    return {
-        "id": provider.id,
-        "provider_type": provider.provider_type,
-        "base_url": provider.base_url,
-        "api_key": "[REDACTED]",
-        "api_version": provider.api_version,
-        "enabled": provider.enabled,
-        "provider_fee": provider.provider_fee,
-        "provider_settings": json.loads(provider.provider_settings)
-        if provider.provider_settings
-        else None,
-    }
+    return _serialize_provider(provider)
+
+
+@admin_router.patch(
+    "/api/upstream-providers", dependencies=[Depends(require_admin_api)]
+)
+async def update_upstream_provider_by_slug(
+    payload: UpstreamProviderUpdateBySlug,
+) -> dict[str, object]:
+    lookup = _validate_slug(payload.slug)
+    async with create_session() as session:
+        result = await session.exec(
+            select(UpstreamProviderRow).where(
+                UpstreamProviderRow.slug == lookup
+            )
+        )
+        provider = result.first()
+        if not provider:
+            raise HTTPException(status_code=404, detail="Provider not found")
+
+        update_payload = UpstreamProviderUpdate(
+            provider_type=payload.provider_type,
+            base_url=payload.base_url,
+            api_key=payload.api_key,
+            api_version=payload.api_version,
+            enabled=payload.enabled,
+            provider_fee=payload.provider_fee,
+            provider_settings=payload.provider_settings,
+        )
+        await _apply_provider_update(
+            session, provider, update_payload, new_slug=payload.new_slug
+        )
+
+    await reinitialize_upstreams()
+    await refresh_model_maps()
+    return _serialize_provider(provider)
 
 
 @admin_router.delete(
     "/api/upstream-providers/{provider_id}", dependencies=[Depends(require_admin_api)]
 )
-async def delete_upstream_provider(provider_id: int) -> dict[str, object]:
+async def delete_upstream_provider(provider_id: str) -> dict[str, object]:
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        deleted_id = _provider_pk(provider)
         await session.delete(provider)
         await session.commit()
     await reinitialize_upstreams()
     await refresh_model_maps()
-    return {"ok": True, "deleted_id": provider_id}
+    return {"ok": True, "deleted_id": deleted_id}
 
 
 @admin_router.get("/api/provider-types", dependencies=[Depends(require_admin_api)])
@@ -885,17 +988,16 @@ async def get_provider_types() -> list[dict[str, object]]:
     "/api/upstream-providers/{provider_id}/models",
     dependencies=[Depends(require_admin_api)],
 )
-async def get_provider_models(provider_id: int) -> dict[str, object]:
+async def get_provider_models(provider_id: str) -> dict[str, object]:
     from ..upstream.helpers import _instantiate_provider
 
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
+        provider_pk = _provider_pk(provider)
 
         db_models = await list_models(
             session=session,
-            upstream_id=provider_id,
+            upstream_id=provider_pk,
             include_disabled=True,
             apply_fees=False,
         )
@@ -985,13 +1087,11 @@ class TopupTokenRequest(BaseModel):
     dependencies=[Depends(require_admin_api)],
 )
 async def topup_provider_with_token(
-    provider_id: int, payload: TopupTokenRequest
+    provider_id: str, payload: TopupTokenRequest
 ) -> dict:
     """Redeem a Cashu token for an upstream provider."""
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
 
         import httpx
 
@@ -1022,15 +1122,13 @@ async def topup_provider_with_token(
     dependencies=[Depends(require_admin_api)],
 )
 async def initiate_provider_topup(
-    provider_id: int, payload: TopupRequest
+    provider_id: str, payload: TopupRequest
 ) -> dict[str, object]:
     """Initiate a Lightning Network top-up for the upstream provider account."""
     from ..upstream.helpers import _instantiate_provider
 
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
 
         try:
             logger.info(
@@ -1150,15 +1248,13 @@ async def initiate_provider_topup(
     "/api/upstream-providers/{provider_id}/topup/{invoice_id}/status",
     dependencies=[Depends(require_admin_api)],
 )
-async def check_topup_status(provider_id: int, invoice_id: str) -> dict[str, object]:
+async def check_topup_status(provider_id: str, invoice_id: str) -> dict[str, object]:
     """Check the status of a Lightning Network top-up invoice."""
     from ..upstream.helpers import _instantiate_provider
     from ..upstream.ppqai import PPQAIUpstreamProvider
 
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
 
         # For Routstr providers, proxy the status check
         if provider.provider_type == "routstr":
@@ -1205,14 +1301,12 @@ async def check_topup_status(provider_id: int, invoice_id: str) -> dict[str, obj
     "/api/upstream-providers/{provider_id}/balance",
     dependencies=[Depends(require_admin_api)],
 )
-async def get_provider_balance(provider_id: int) -> dict[str, object]:
+async def get_provider_balance(provider_id: str) -> dict[str, object]:
     """Get the current balance for an upstream provider account."""
     from ..upstream.helpers import _instantiate_provider
 
     async with create_session() as session:
-        provider = await session.get(UpstreamProviderRow, provider_id)
-        if not provider:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider = await _get_upstream_provider_by_ref(session, provider_id)
 
         # For Routstr providers, proxy the balance check
         if provider.provider_type == "routstr":
@@ -1591,15 +1685,13 @@ async def get_lightning_invoices_api(
     "/api/upstream-providers/{provider_id}/routstr/refund",
     dependencies=[Depends(require_admin_api)],
 )
-async def refund_routstr_provider_balance(provider_id: int) -> dict[str, object]:
+async def refund_routstr_provider_balance(provider_id: str) -> dict[str, object]:
     """Refund balance from an upstream Routstr provider back to the local wallet."""
     from ..upstream.helpers import _instantiate_provider
     from ..upstream.routstr import RoutstrUpstreamProvider
 
     async with create_session() as session:
-        provider_row = await session.get(UpstreamProviderRow, provider_id)
-        if not provider_row:
-            raise HTTPException(status_code=404, detail="Provider not found")
+        provider_row = await _get_upstream_provider_by_ref(session, provider_id)
 
         if provider_row.provider_type != "routstr":
             raise HTTPException(

routstr/core/db.py

@@ -9,9 +9,10 @@ from typing import AsyncGenerator
 from alembic import command
 from alembic.config import Config
 from alembic.util.exc import CommandError
-from sqlalchemy import UniqueConstraint
+from sqlalchemy import UniqueConstraint, delete
 from sqlalchemy.exc import OperationalError
 from sqlalchemy.ext.asyncio.engine import create_async_engine
+from sqlalchemy.orm import aliased
 from sqlmodel import Field, Relationship, SQLModel, col, func, select, update
 from sqlmodel.ext.asyncio.session import AsyncSession
 
@@ -125,6 +126,63 @@ async def release_stale_reservations(
     return released
 
 
+async def prune_dead_api_keys(session: AsyncSession, min_age_seconds: int) -> int:
+    """Delete dead parentless API keys; return the count removed.
+
+    Dead = 0 balance/reservation/spend/requests, older than the grace period,
+    no parent, no children, no pending invoice. Cashu rows are unlinked (not
+    deleted) first to keep the audit trail.
+    """
+    cutoff = int(time.time()) - min_age_seconds
+
+    child = aliased(ApiKey)
+    has_children = (
+        select(child.hashed_key).where(
+            col(child.parent_key_hash) == col(ApiKey.hashed_key)
+        )
+    ).exists()
+    pending_invoice = (
+        select(LightningInvoice.id)
+        .where(col(LightningInvoice.api_key_hash) == col(ApiKey.hashed_key))
+        .where(col(LightningInvoice.status) == "pending")
+    ).exists()
+
+    eligible_hashes = (
+        select(ApiKey.hashed_key)
+        .where(col(ApiKey.balance) == 0)
+        .where(col(ApiKey.reserved_balance) == 0)
+        .where(col(ApiKey.total_spent) == 0)
+        .where(col(ApiKey.total_requests) == 0)
+        .where(col(ApiKey.parent_key_hash).is_(None))
+        .where(
+            (col(ApiKey.created_at).is_(None)) | (col(ApiKey.created_at) < cutoff)
+        )
+        .where(~pending_invoice)
+        .where(~has_children)
+    )
+
+    # Unlink transactions rather than cascade-deleting them, so the financial
+    # audit trail survives. The eligibility predicate is re-evaluated inside both
+    # statements so a key that gained balance mid-run is left untouched.
+    await session.exec(  # type: ignore[call-overload]
+        update(CashuTransaction)
+        .where(col(CashuTransaction.api_key_hashed_key).in_(eligible_hashes))
+        .values(api_key_hashed_key=None)
+    )
+
+    result = await session.exec(  # type: ignore[call-overload]
+        delete(ApiKey).where(col(ApiKey.hashed_key).in_(eligible_hashes))
+    )
+    await session.commit()
+
+    pruned = int(result.rowcount or 0)
+    logger.info(
+        "Pruned dead API keys",
+        extra={"pruned_keys": pruned, "min_age_seconds": min_age_seconds},
+    )
+    return pruned
+
+
 class ModelRow(SQLModel, table=True):  # type: ignore
     __tablename__ = "models"
     id: str = Field(primary_key=True)
@@ -261,6 +319,12 @@ class UpstreamProviderRow(SQLModel, table=True):  # type: ignore
         ),
     )
     id: int | None = Field(default=None, primary_key=True)
+    slug: str | None = Field(
+        default=None,
+        unique=True,
+        index=True,
+        description="Stable external slug used for updates via API key.",
+    )
     provider_type: str = Field(
         description="Provider type: custom, openai, anthropic, azure, openrouter, etc."
     )

routstr/core/exceptions.py

@@ -7,11 +7,26 @@ logger = get_logger(__name__)
 
 
 class UpstreamError(Exception):
-    """Exception raised when an upstream provider fails."""
+    """Exception raised when an upstream provider fails.
 
-    def __init__(self, message: str, status_code: int = 502):
+    ``code`` carries a stable, machine-readable classification (e.g.
+    ``UPSTREAM_RATE_LIMIT``) so callers can distinguish failure kinds without
+    string-matching the message. ``details`` holds optional structured,
+    redaction-safe context. Both default to ``None`` for backwards
+    compatibility.
+    """
+
+    def __init__(
+        self,
+        message: str,
+        status_code: int = 502,
+        code: str | None = None,
+        details: dict[str, object] | None = None,
+    ):
         self.message = message
         self.status_code = status_code
+        self.code = code
+        self.details = details
         super().__init__(message)
 
 

routstr/core/logging.py

@@ -51,6 +51,8 @@ from pythonjsonlogger import jsonlogger
 from rich.console import Console
 from rich.logging import RichHandler
 
+from .redaction import redact_obj, redact_org_ids
+
 # Only use RichHandler when stdout is a real TTY. In non-TTY contexts
 # (docker logs, pipes, CI) Rich pads every line to width and wraps long
 # records, producing visually-empty trailing whitespace and split records.
@@ -180,6 +182,37 @@ class RequestIdFilter(logging.Filter):
         return True
 
 
+# Standard ``LogRecord`` attributes that are never user-supplied ``extra``
+# fields; skipped when redacting structured extras (``msg``/``message`` are
+# handled separately above).
+_NON_EXTRA_RECORD_ATTRS = frozenset(
+    {
+        "name",
+        "msg",
+        "args",
+        "levelname",
+        "levelno",
+        "pathname",
+        "filename",
+        "module",
+        "exc_info",
+        "exc_text",
+        "stack_info",
+        "lineno",
+        "funcName",
+        "created",
+        "msecs",
+        "relativeCreated",
+        "thread",
+        "threadName",
+        "processName",
+        "process",
+        "taskName",
+        "message",
+    }
+)
+
+
 class SecurityFilter(logging.Filter):
     """Filter to remove sensitive information from logs."""
 
@@ -203,6 +236,7 @@ class SecurityFilter(logging.Filter):
         """Filter out sensitive information from log records."""
         try:
             message = record.getMessage()
+            message = redact_org_ids(message)
             standalone_patterns = [
                 r"Bearer\s+([a-zA-Z0-9_\-\.]{10,})",  # Bearer token (must be 10 characters or more to reduce false-positives)
                 r"cashu[A-Z]+([a-zA-Z0-9_\-\.=/+]+)",  # Cashu tokens
@@ -224,6 +258,16 @@ class SecurityFilter(logging.Filter):
             record.msg = message
             record.args = ()
 
+            # Structured `extra={...}` fields are emitted by the JSON formatter
+            # straight from the record dict and never pass through the message
+            # formatting above. Redact organization IDs from any string-valued
+            # extra so they cannot leak via structured logs.
+            for attr, value in list(record.__dict__.items()):
+                if attr in _NON_EXTRA_RECORD_ATTRS:
+                    continue
+                if isinstance(value, (str, dict, list, tuple)):
+                    record.__dict__[attr] = redact_obj(value)
+
         except Exception:
             pass
 

routstr/core/main.py

@@ -11,7 +11,11 @@ from starlette.exceptions import HTTPException
 from starlette.responses import Response as StarletteResponse
 from starlette.types import Scope
 
-from ..auth import periodic_key_reset, periodic_stale_reservation_sweep
+from ..auth import (
+    periodic_dead_key_prune,
+    periodic_key_reset,
+    periodic_stale_reservation_sweep,
+)
 from ..balance import balance_router, deprecated_wallet_router
 from ..lightning import lightning_router, periodic_invoice_watcher
 from ..nostr import (
@@ -56,6 +60,7 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
     model_maps_refresh_task = None
     key_reset_task = None
     stale_reservation_task = None
+    dead_key_prune_task = None
     auto_topup_task = None
     refund_sweep_task = None
     routstr_fee_task = None
@@ -132,6 +137,7 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
         stale_reservation_task = asyncio.create_task(
             periodic_stale_reservation_sweep()
         )
+        dead_key_prune_task = asyncio.create_task(periodic_dead_key_prune())
         auto_topup_task = asyncio.create_task(periodic_auto_topup())
         refund_sweep_task = asyncio.create_task(periodic_refund_sweep())
         routstr_fee_task = asyncio.create_task(periodic_routstr_fee_payout())
@@ -171,6 +177,8 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
             key_reset_task.cancel()
         if stale_reservation_task is not None:
             stale_reservation_task.cancel()
+        if dead_key_prune_task is not None:
+            dead_key_prune_task.cancel()
         if auto_topup_task is not None:
             auto_topup_task.cancel()
         if refund_sweep_task is not None:
@@ -202,6 +210,8 @@ async def lifespan(_: FastAPI) -> AsyncGenerator[None, None]:
                 tasks_to_wait.append(key_reset_task)
             if stale_reservation_task is not None:
                 tasks_to_wait.append(stale_reservation_task)
+            if dead_key_prune_task is not None:
+                tasks_to_wait.append(dead_key_prune_task)
             if auto_topup_task is not None:
                 tasks_to_wait.append(auto_topup_task)
             if refund_sweep_task is not None:
@@ -369,7 +379,10 @@ if UI_DIST_PATH.exists() and UI_DIST_PATH.is_dir():
 
 else:
     logger.warning(
-        f"UI dist directory not found at {UI_DIST_PATH}, skipping static file serving"
+        "UI dist directory not found at %s; serving API only. Run `make ui-build` "
+        "to build the static UI served from here, or `make ui-dev` for the Next.js "
+        "dev server with hot reload on :3000 (it targets this backend on :8000).",
+        UI_DIST_PATH,
     )
 
     @app.get("/", include_in_schema=False)

routstr/core/provider_slugs.py

@@ -0,0 +1,65 @@
+from __future__ import annotations
+
+import re
+from itertools import count
+from typing import Collection
+
+from sqlmodel import select
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from .db import UpstreamProviderRow
+
+_SLUG_BASE_PATTERN = re.compile(r"[^a-z0-9]+")
+_MAX_SLUG_LENGTH = 64
+
+
+def provider_slug_base(provider_type: str) -> str:
+    """Return a deterministic slug base for a provider type."""
+    base = _SLUG_BASE_PATTERN.sub("-", provider_type.lower()).strip("-")
+    if not base:
+        base = "provider"
+    elif base.isdigit():
+        base = f"provider-{base}"
+    elif len(base) < 3:
+        base = f"{base}-provider"
+
+    if len(base) > _MAX_SLUG_LENGTH:
+        base = base[:_MAX_SLUG_LENGTH].rstrip("-") or "provider"
+    return base
+
+
+def provider_slug_candidate(base: str, suffix_number: int) -> str:
+    if suffix_number == 1:
+        return base
+
+    suffix = f"-{suffix_number}"
+    max_base_length = _MAX_SLUG_LENGTH - len(suffix)
+    return f"{base[:max_base_length].rstrip('-')}{suffix}"
+
+
+async def allocate_unique_provider_slug(
+    session: AsyncSession,
+    provider_type: str,
+    reserved_slugs: Collection[str] = (),
+) -> str:
+    """Allocate a stable, deterministic provider slug.
+
+    The first provider of a type gets ``openai``; later collisions get
+    ``openai-2``, ``openai-3``, etc. ``reserved_slugs`` covers rows staged in
+    memory but not flushed yet, such as settings/env seeding.
+    """
+    base = provider_slug_base(provider_type)
+    reserved = {slug.lower() for slug in reserved_slugs}
+
+    for suffix_number in count(1):
+        candidate = provider_slug_candidate(base, suffix_number)
+        if candidate in reserved:
+            continue
+
+        result = await session.exec(
+            select(UpstreamProviderRow).where(UpstreamProviderRow.slug == candidate)
+        )
+        if result.first() is None:
+            return candidate
+
+    raise RuntimeError("unreachable")

routstr/core/redaction.py

@@ -0,0 +1,51 @@
+"""Redaction helpers for sensitive provider identifiers.
+
+Single source of truth for stripping account-scoped identifiers (e.g. OpenAI
+organization IDs) from any text before it is logged, returned to a caller, or
+written to an audit entry.
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Any
+
+# OpenAI-style organization identifiers look like ``org-<base62>``. Require at
+# least 6 trailing chars so the already-redacted literal ``org-[REDACTED]`` is
+# never re-matched (``[`` is not in the character class).
+_ORG_ID_PATTERN = re.compile(r"\borg-[A-Za-z0-9]{6,}\b")
+
+ORG_ID_PLACEHOLDER = "org-[REDACTED]"
+
+
+def redact_org_ids(text: str) -> str:
+    """Replace OpenAI-style organization IDs with ``org-[REDACTED]``.
+
+    Args:
+        text: Arbitrary text that may embed an ``org-*`` identifier.
+
+    Returns:
+        The text with every organization ID replaced. Non-string input is
+        returned unchanged after coercion to ``str``.
+    """
+    if not text:
+        return text
+    return _ORG_ID_PATTERN.sub(ORG_ID_PLACEHOLDER, text)
+
+
+def redact_obj(obj: Any) -> Any:
+    """Recursively redact organization IDs in arbitrary nested structures.
+
+    Strings are redacted in place; dicts and lists/tuples are walked so that
+    identifiers nested inside structured payloads (e.g. log ``extra`` fields or
+    error ``details``) are also stripped. Other types are returned unchanged.
+    """
+    if isinstance(obj, str):
+        return redact_org_ids(obj)
+    if isinstance(obj, dict):
+        return {key: redact_obj(value) for key, value in obj.items()}
+    if isinstance(obj, list):
+        return [redact_obj(value) for value in obj]
+    if isinstance(obj, tuple):
+        return tuple(redact_obj(value) for value in obj)
+    return obj

routstr/core/settings.py

@@ -73,6 +73,14 @@ class Settings(BaseSettings):
     stale_reservation_timeout_seconds: int = Field(
         default=300, env="STALE_RESERVATION_TIMEOUT_SECONDS"
     )
+    # Background prune of dead (zero balance, never used) API keys.
+    # Interval 0 disables it; min-age is a grace period (default 1 week).
+    dead_key_prune_interval_seconds: int = Field(
+        default=3600, env="DEAD_KEY_PRUNE_INTERVAL_SECONDS"
+    )
+    dead_key_min_age_seconds: int = Field(
+        default=604_800, env="DEAD_KEY_MIN_AGE_SECONDS"
+    )
 
     # Network
     cors_origins: list[str] = Field(default_factory=lambda: ["*"], env="CORS_ORIGINS")

routstr/core/version.py

@@ -20,7 +20,7 @@ import subprocess
 from functools import lru_cache
 from pathlib import Path
 
-BASE_VERSION = "0.4.3"
+BASE_VERSION = "0.4.4"
 
 _REPO_ROOT = Path(__file__).resolve().parents[2]
 _GIT_TIMEOUT_SECONDS = 2.0

routstr/payment/cost_calculation.py

@@ -41,6 +41,28 @@ class CostDataError(BaseModel):
     code: str
 
 
+def _empty_cost(cls: type[CostData] = CostData) -> CostData:
+    """Build an all-zero cost object — a full refund for an empty response.
+
+    Shared by the two paths that must not bill: an upstream response with no
+    usage data at all, and one that reports a USD cost but carries zero tokens
+    in every bucket.
+    """
+    return cls(
+        base_msats=0,
+        input_msats=0,
+        output_msats=0,
+        total_msats=0,
+        total_usd=0.0,
+        input_tokens=0,
+        output_tokens=0,
+        cache_read_input_tokens=0,
+        cache_creation_input_tokens=0,
+        cache_read_msats=0,
+        cache_creation_msats=0,
+    )
+
+
 async def calculate_cost(
     response_data: dict,
     max_cost: int,
@@ -83,19 +105,7 @@ async def calculate_cost(
                 else None,
             },
         )
-        return MaxCostData(
-            base_msats=0,
-            input_msats=0,
-            output_msats=0,
-            total_msats=0,
-            total_usd=0.0,
-            input_tokens=0,
-            output_tokens=0,
-            cache_read_input_tokens=0,
-            cache_creation_input_tokens=0,
-            cache_read_msats=0,
-            cache_creation_msats=0,
-        )
+        return _empty_cost(MaxCostData)
 
     usage_data = response_data.get("usage") or {}
     if not isinstance(usage_data, dict):
@@ -109,6 +119,27 @@ async def calculate_cost(
     # Try USD cost first
     usd_cost = _resolve_usd_cost(usage_data, response_data)
     if usd_cost > 0:
+        truly_empty = (
+            input_tokens == 0
+            and output_tokens == 0
+            and cache_read_tokens == 0
+            and cache_creation_tokens == 0
+        )
+        if truly_empty:
+            logger.warning(
+                "Upstream reported a USD cost but the response carries no "
+                "tokens at all (input, output, cache-read and cache-creation "
+                "are all zero) — refunding in full rather than billing the "
+                "USD-derived cost for an empty response.",
+                extra={
+                    "model": response_data.get("model", "unknown"),
+                    "usd_cost": usd_cost,
+                    "usage_keys": sorted(usage_data.keys())
+                    if isinstance(usage_data, dict)
+                    else None,
+                },
+            )
+            return _empty_cost()
         if input_tokens == 0 and output_tokens == 0:
             logger.warning(
                 "Upstream reported a USD cost but no token counts — "

routstr/payment/helpers.py

@@ -11,6 +11,8 @@ from PIL import Image
 from sqlmodel.ext.asyncio.session import AsyncSession
 
 from ..core import get_logger
+from ..core.exceptions import UpstreamError
+from ..core.redaction import redact_org_ids
 from ..core.settings import settings
 from ..wallet import deserialize_token_from_string
 
@@ -418,16 +420,27 @@ def create_error_response(
     status_code: int,
     request: Request,
     token: str | None = None,
+    code: str | int | None = None,
+    details: dict[str, object] | None = None,
 ) -> Response:
-    """Create a standardized error response."""
+    """Create a standardized error response.
+
+    ``code`` is a stable, machine-readable classification (e.g.
+    ``UPSTREAM_RATE_LIMIT``); when omitted it defaults to the HTTP status code
+    for backwards compatibility. ``details`` carries optional structured,
+    redaction-safe context.
+    """
+    error_obj: dict[str, object] = {
+        "message": redact_org_ids(message),
+        "type": error_type,
+        "code": code if code is not None else status_code,
+    }
+    if details is not None:
+        error_obj["details"] = details
     return Response(
         content=json.dumps(
             {
-                "error": {
-                    "message": message,
-                    "type": error_type,
-                    "code": status_code,
-                },
+                "error": error_obj,
                 "request_id": getattr(request.state, "request_id", "unknown"),
             }
         ),
@@ -435,3 +448,20 @@ def create_error_response(
         media_type="application/json",
         headers={"X-Cashu": token} if token else {},
     )
+
+
+def create_upstream_error_response(
+    error: UpstreamError,
+    request: Request,
+    fallback_status: int = 502,
+) -> Response:
+    """Build an error response from an :class:`UpstreamError`, preserving its
+    structured ``code``, ``details``, and original ``status_code``."""
+    return create_error_response(
+        "upstream_error",
+        str(error),
+        error.status_code or fallback_status,
+        request=request,
+        code=getattr(error, "code", None),
+        details=getattr(error, "details", None),
+    )

routstr/payment/models.py

@@ -85,6 +85,30 @@ class Model(BaseModel):
         return hash(self.id)
 
 
+def litellm_cost_entry(model_id: str) -> dict | None:
+    """Look up ``model_id`` in litellm's bundled cost map.
+
+    litellm ships per-model USD rates keyed by the exact OpenRouter id
+    (``deepseek/deepseek-chat``) or the bare model name (``gpt-4o``,
+    ``claude-sonnet-4-5``), so both spellings are tried. Keys are lowercase, so
+    a mixed-case upstream id (``deepseek-ai/DeepSeek-V4-Flash``) is retried via
+    a case-insensitive scan. Returns the matched cost dict, or ``None``.
+    """
+    import litellm
+
+    candidates = (model_id, model_id.split("/", 1)[-1])
+    for key in candidates:
+        info = litellm.model_cost.get(key)
+        if isinstance(info, dict):
+            return info
+
+    lowered = {c.lower() for c in candidates}
+    for key, info in litellm.model_cost.items():
+        if isinstance(key, str) and key.lower() in lowered and isinstance(info, dict):
+            return info
+    return None
+
+
 def backfill_cache_pricing(model_id: str, pricing: Pricing) -> Pricing:
     """Fill missing cache rates from litellm's bundled cost map.
 
@@ -92,9 +116,8 @@ def backfill_cache_pricing(model_id: str, pricing: Pricing) -> Pricing:
     for many models (most DeepSeek entries, openai/gpt-4o, ...). Without a
     cache rate, billing falls back to the full input rate, which overcharges
     cache reads (DeepSeek hits are 10x cheaper) and undercharges Anthropic
-    cache writes (1.25x). litellm ships per-model USD rates keyed by the exact
-    OpenRouter id (deepseek/deepseek-chat) or by the bare model name
-    (gpt-4o, claude-sonnet-4-5), so both spellings are tried.
+    cache writes (1.25x). The lookup (see ``litellm_cost_entry``) tries both
+    id spellings and a case-insensitive fallback.
 
     Rates already present (e.g. provided by OpenRouter) are authoritative and
     never overwritten. Unknown models are returned unchanged.
@@ -104,14 +127,7 @@ def backfill_cache_pricing(model_id: str, pricing: Pricing) -> Pricing:
     if not (needs_read or needs_write):
         return pricing
 
-    import litellm
-
-    info: dict | None = None
-    for key in (model_id, model_id.split("/", 1)[-1]):
-        candidate = litellm.model_cost.get(key)
-        if isinstance(candidate, dict):
-            info = candidate
-            break
+    info = litellm_cost_entry(model_id)
     if info is None:
         return pricing
 
@@ -215,13 +231,29 @@ def _row_to_model(
     )
     top_provider_dict = json.loads(row.top_provider) if row.top_provider else None
 
-    if apply_provider_fee and isinstance(pricing, dict):
-        pricing = {k: float(v) * provider_fee for k, v in pricing.items()}
-
     if isinstance(pricing, dict) and float(pricing.get("request", 0.0)) <= 0.0:
         pricing["request"] = max(pricing.get("request", 0.0), 0.0)
 
     parsed_pricing = Pricing.parse_obj(pricing)
+
+    # Fill missing cache-read/write rates from litellm's cost map BEFORE applying
+    # the provider fee, so they carry the same markup as every other component.
+    # DB-stored override pricing (e.g. generic providers) omits cache rates;
+    # without this, ``_row_to_model`` bills cache reads at the full input rate —
+    # the ``_apply_provider_fee_to_model`` path backfills, but the override path
+    # used for admin-configured providers did not.
+    #
+    # Key on ``forwarded_model_id`` (the actual upstream model name litellm
+    # prices) when set: an alias row (id="local-alias",
+    # forwarded_model_id="deepseek-v4-flash") would otherwise look up the alias
+    # and miss the cache rate.
+    pricing_model_id = getattr(row, "forwarded_model_id", None) or row.id
+    parsed_pricing = backfill_cache_pricing(pricing_model_id, parsed_pricing)
+
+    if apply_provider_fee:
+        parsed_pricing = Pricing.parse_obj(
+            {k: float(v) * provider_fee for k, v in parsed_pricing.dict().items()}
+        )
     model = Model(
         id=row.id,
         name=row.name,

routstr/proxy.py

@@ -24,6 +24,7 @@ from .payment.helpers import (
     calculate_discounted_max_cost,
     check_token_balance,
     create_error_response,
+    create_upstream_error_response,
     get_max_cost_for_model,
 )
 from .payment.models import Model
@@ -211,9 +212,7 @@ async def proxy(
                     e,
                 )
                 if i == len(all_upstreams) - 1:
-                    last_error_response = create_error_response(
-                        "upstream_error", str(e), 502, request=request
-                    )
+                    last_error_response = create_upstream_error_response(e, request)
                 continue
         return last_error_response or create_error_response(
             "upstream_error", "All upstreams failed", 502, request=request
@@ -283,11 +282,10 @@ async def proxy(
                     last_error = e
                 continue
 
+        if last_error is not None:
+            return create_upstream_error_response(last_error, request)
         return create_error_response(
-            "upstream_error",
-            str(last_error) if last_error else "All upstreams failed",
-            502,
-            request=request,
+            "upstream_error", "All upstreams failed", 502, request=request
         )
 
     elif auth := headers.get("authorization", None):
@@ -343,9 +341,7 @@ async def proxy(
             except UpstreamError as e:
                 logger.warning(f"Upstream {upstream.provider_type} failed (GET): {e}")
                 if i == len(upstreams) - 1:
-                    last_error_response = create_error_response(
-                        "upstream_error", str(e), 502, request=request
-                    )
+                    last_error_response = create_upstream_error_response(e, request)
                 continue
         return last_error_response or create_error_response(
             "upstream_error", "All upstreams failed", 502, request=request
@@ -525,9 +521,7 @@ async def proxy(
             # If this was the last provider
             if i == len(upstreams) - 1:
                 await revert_pay_for_request(key, session, max_cost_for_model)
-                return create_error_response(
-                    "upstream_error", str(e), 502, request=request
-                )
+                return create_upstream_error_response(e, request)
 
             # Otherwise loop continues to next provider
             continue

routstr/upstream/anthropic.py

@@ -24,7 +24,7 @@ class AnthropicUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "AnthropicUpstreamProvider":
         return cls(

routstr/upstream/auto_topup.py

@@ -97,6 +97,8 @@ async def _check_and_topup(row: UpstreamProviderRow) -> None:
 
     # Instantiate provider and check balance
     provider = RoutstrUpstreamProvider.from_db_row(row)
+    if provider is None:
+        return
     balance = await provider.get_balance()
 
     if balance is None:

routstr/upstream/azure.py

@@ -38,7 +38,7 @@ class AzureUpstreamProvider(BaseUpstreamProvider):
         self.api_version = api_version
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "AzureUpstreamProvider | None":
         if not provider_row.api_version:

routstr/upstream/base.py

@@ -2,16 +2,16 @@ from __future__ import annotations
 
 import asyncio
 import json
+import math
 import traceback
 import uuid
 from collections.abc import AsyncGenerator, AsyncIterator, Iterator
-from typing import Any, Mapping, cast
+from typing import Any, Mapping, Self, cast
 
 import httpx
 from fastapi import BackgroundTasks, HTTPException, Request
 from fastapi.responses import Response, StreamingResponse
 from pydantic.v1 import BaseModel
-from sqlmodel import select
 
 from ..auth import adjust_payment_for_tokens
 from ..core import get_logger
@@ -23,6 +23,7 @@ from ..core.db import (
     store_cashu_transaction,
 )
 from ..core.exceptions import UpstreamError
+from ..core.redaction import redact_org_ids
 from ..payment.cost_calculation import (
     CostData,
     CostDataError,
@@ -47,6 +48,7 @@ from .cache_breakpoints import (
 )
 from .count_tokens import count_tokens_locally
 from .litellm_routing import detect_litellm_prefix
+from .rate_limit import UPSTREAM_RATE_LIMIT, classify_rate_limit
 
 logger = get_logger(__name__)
 
@@ -89,6 +91,11 @@ class BaseUpstreamProvider:
     base_url: str
     api_key: str
     provider_fee: float = 1.05
+    # Primary key of the ``upstream_providers`` row this instance was built
+    # from. Set by ``from_db_row`` so a live provider can re-find its own row by
+    # stable identity instead of its rotatable ``api_key``. ``None`` for
+    # instances not sourced from a row.
+    db_id: int | None = None
     _models_cache: list[Model] = []
     _models_by_id: dict[str, Model] = {}
 
@@ -103,6 +110,7 @@ class BaseUpstreamProvider:
         self.base_url = base_url
         self.api_key = api_key
         self.provider_fee = provider_fee
+        self.db_id = None
         self._models_cache = []
         self._models_by_id = {}
 
@@ -120,10 +128,13 @@ class BaseUpstreamProvider:
         return detect_litellm_prefix(self.base_url)
 
     @classmethod
-    def from_db_row(
-        cls, provider_row: "UpstreamProviderRow"
-    ) -> "BaseUpstreamProvider | None":
-        """Factory method to instantiate provider from database row.
+    def from_db_row(cls, provider_row: "UpstreamProviderRow") -> "Self | None":
+        """Instantiate a provider from a database row, carrying its identity.
+
+        Construction itself is delegated to the ``_build_from_row`` hook (which
+        subclasses override to match their constructor); this wrapper stamps the
+        row's primary key onto the instance as ``db_id`` so the provider can
+        later re-find its own row by identity rather than by its ``api_key``.
 
         Args:
             provider_row: Database row containing provider configuration
@@ -131,6 +142,19 @@ class BaseUpstreamProvider:
         Returns:
             Instantiated provider or None if instantiation fails
         """
+        provider = cls._build_from_row(provider_row)
+        if provider is not None:
+            provider.db_id = provider_row.id
+        return provider
+
+    @classmethod
+    def _build_from_row(cls, provider_row: "UpstreamProviderRow") -> "Self | None":
+        """Construct the provider instance from a row (no identity stamping).
+
+        Overridden by subclasses whose constructors differ from the base
+        ``(base_url, api_key, provider_fee)`` shape. Callers should use
+        ``from_db_row`` instead, which also attaches ``db_id``.
+        """
         return cls(
             base_url=provider_row.base_url,
             api_key=provider_row.api_key,
@@ -582,7 +606,7 @@ class BaseUpstreamProvider:
             preview = body_bytes.decode("utf-8", errors="ignore").strip()
             if preview:
                 message = preview[:500]
-        return message, upstream_code
+        return redact_org_ids(message), upstream_code
 
     async def on_upstream_error_redirect(
         self, status_code: int, error_message: str
@@ -625,10 +649,23 @@ class BaseUpstreamProvider:
             body_bytes = b""
             body_read_error = f"{type(exc).__name__}: {exc}"
 
+        # ``message`` is already redacted by ``_extract_upstream_error_message``;
+        # the raw body preview is redacted here before it reaches logs or the
+        # forwarded envelope so provider account identifiers never leak.
         message, upstream_code = self._extract_upstream_error_message(body_bytes)
-        body_preview = body_bytes.decode("utf-8", errors="ignore").strip()[:500]
+        body_preview = redact_org_ids(
+            body_bytes.decode("utf-8", errors="ignore").strip()[:500]
+        )
         is_json_body = _is_json_content_type(content_type)
 
+        # Classify upstream rate-limit failures into a stable, structured error.
+        rate_limit = classify_rate_limit(status_code, message, headers)
+        error_code: str | int = upstream_code or status_code
+        error_details: dict[str, object] | None = None
+        if rate_limit is not None:
+            error_code = UPSTREAM_RATE_LIMIT
+            error_details = rate_limit.as_details()
+
         logger.warning(
             "Upstream %s returned %s for model=%s path=%s: %s",
             self.provider_type,
@@ -642,6 +679,7 @@ class BaseUpstreamProvider:
                 "model": model_id or "unknown",
                 "upstream_status": status_code,
                 "upstream_code": upstream_code,
+                "error_code": error_code,
                 "upstream_content_type": content_type,
                 "upstream_request_id": upstream_request_id,
                 "message_preview": message[:200],
@@ -676,13 +714,41 @@ class BaseUpstreamProvider:
         ):
             headers.pop(header_name, None)
 
+        # Propagate a usable retry hint to the caller when the upstream supplied
+        # one but did not echo a ``Retry-After`` header. RFC 7231 delta-seconds
+        # is an integer, so round sub-second hints up to a usable ``1``.
+        if (
+            rate_limit is not None
+            and rate_limit.retry_after_seconds is not None
+            and "retry-after" not in {k.lower() for k in headers}
+        ):
+            headers["Retry-After"] = str(max(1, math.ceil(rate_limit.retry_after_seconds)))
+
         if is_json_body:
             if not content_type:
                 headers.pop("content-type", None)
                 headers.pop("Content-Type", None)
             media_type = content_type or None
+            # Re-serialise the body with organization IDs stripped. The narrow
+            # ``org-*`` regex preserves the surrounding JSON structure.
+            redacted_text = redact_org_ids(body_bytes.decode("utf-8", errors="ignore"))
+            redacted_body = redacted_text.encode()
+            # Surface the stable rate-limit classification on the forwarded
+            # body so callers can switch on ``error.code`` without parsing the
+            # provider-specific message. Fall back to the redacted bytes if the
+            # body is not a JSON object with an ``error`` mapping.
+            if rate_limit is not None:
+                try:
+                    parsed = json.loads(redacted_text)
+                    err = parsed.get("error") if isinstance(parsed, dict) else None
+                    if isinstance(err, dict):
+                        err["code"] = UPSTREAM_RATE_LIMIT
+                        err["details"] = error_details
+                        redacted_body = json.dumps(parsed).encode()
+                except (ValueError, AttributeError):
+                    pass
             return Response(
-                content=body_bytes,
+                content=redacted_body,
                 status_code=status_code,
                 headers=headers,
                 media_type=media_type,
@@ -693,15 +759,18 @@ class BaseUpstreamProvider:
         for header_name in ("content-type", "Content-Type"):
             headers.pop(header_name, None)
 
+        error_obj: dict[str, object] = {
+            "message": message or "Upstream returned a non-JSON error response",
+            "type": "upstream_error",
+            "code": error_code,
+            "upstream_status": status_code,
+            "upstream_content_type": content_type or None,
+            "upstream_body_preview": body_preview or None,
+        }
+        if error_details is not None:
+            error_obj["details"] = error_details
         envelope = {
-            "error": {
-                "message": message or "Upstream returned a non-JSON error response",
-                "type": "upstream_error",
-                "code": upstream_code or status_code,
-                "upstream_status": status_code,
-                "upstream_content_type": content_type or None,
-                "upstream_body_preview": body_preview or None,
-            },
+            "error": error_obj,
             "request_id": getattr(request.state, "request_id", None),
         }
 
@@ -2521,9 +2590,16 @@ class BaseUpstreamProvider:
                         body_bytes = await response.aread()
                     except Exception:
                         body_bytes = b""
-                    body_preview = body_bytes.decode(
-                        "utf-8", errors="ignore"
-                    ).strip()[:500]
+                    # Redact provider account identifiers before the body text
+                    # reaches logs or the raised error.
+                    body_preview = redact_org_ids(
+                        body_bytes.decode("utf-8", errors="ignore").strip()[:500]
+                    )
+                    rate_limit = classify_rate_limit(
+                        response.status_code,
+                        body_preview,
+                        dict(response.headers),
+                    )
                     logger.error(
                         "Upstream %s returned %s for model=%s path=%s: %s",
                         self.provider_type,
@@ -2535,6 +2611,7 @@ class BaseUpstreamProvider:
                             "provider": self.provider_type,
                             "model": original_model_id or "unknown",
                             "status_code": response.status_code,
+                            "error_code": rate_limit.code if rate_limit else None,
                             "reason_phrase": response.reason_phrase,
                             "path": path,
                             "body_preview": body_preview,
@@ -2547,6 +2624,8 @@ class BaseUpstreamProvider:
                         f"for model {original_model_id or 'unknown'}: "
                         f"{body_preview[:200] or '<empty>'}",
                         status_code=response.status_code,
+                        code=rate_limit.code if rate_limit else None,
+                        details=rate_limit.as_details() if rate_limit else None,
                     )
 
                 try:
@@ -2843,9 +2922,16 @@ class BaseUpstreamProvider:
                         body_bytes = await response.aread()
                     except Exception:
                         body_bytes = b""
-                    body_preview = body_bytes.decode(
-                        "utf-8", errors="ignore"
-                    ).strip()[:500]
+                    # Redact provider account identifiers before the body text
+                    # reaches logs or the raised error.
+                    body_preview = redact_org_ids(
+                        body_bytes.decode("utf-8", errors="ignore").strip()[:500]
+                    )
+                    rate_limit = classify_rate_limit(
+                        response.status_code,
+                        body_preview,
+                        dict(response.headers),
+                    )
                     logger.error(
                         "Upstream %s returned %s for model=%s path=%s: %s",
                         self.provider_type,
@@ -2857,6 +2943,7 @@ class BaseUpstreamProvider:
                             "provider": self.provider_type,
                             "model": original_model_id or "unknown",
                             "status_code": response.status_code,
+                            "error_code": rate_limit.code if rate_limit else None,
                             "path": path,
                             "body_preview": body_preview,
                         },
@@ -2868,6 +2955,8 @@ class BaseUpstreamProvider:
                         f"for model {original_model_id or 'unknown'}: "
                         f"{body_preview[:200] or '<empty>'}",
                         status_code=response.status_code,
+                        code=rate_limit.code if rate_limit else None,
+                        details=rate_limit.as_details() if rate_limit else None,
                     )
 
                 try:
@@ -4810,14 +4899,11 @@ class BaseUpstreamProvider:
         """Refresh the in-memory models cache from upstream API."""
         try:
             async with create_session() as session:
-                stmt = select(UpstreamProviderRow).where(
-                    UpstreamProviderRow.base_url == self.base_url,
-                    UpstreamProviderRow.api_key == self.api_key,
+                provider = (
+                    await session.get(UpstreamProviderRow, self.db_id)
+                    if self.db_id is not None
+                    else None
                 )
-                result = await session.exec(stmt)
-
-                # .first() returns the object or None if not found
-                provider = result.first()
                 if not provider or not provider.id:
                     raise HTTPException(status_code=404, detail="Provider not found")
 

routstr/upstream/deepseek_v4_pricing_shim.py

@@ -3,10 +3,15 @@
 litellm's bundled cost map does not yet ship ``deepseek-v4-flash`` /
 ``deepseek-v4-pro``. Without an entry, ``backfill_cache_pricing`` cannot find a
 ``cache_read_input_token_cost`` and cache reads fall back to the full input
-rate — a ~60% overcharge on cache hits (DeepSeek hits are ~0.2x input).
+rate — a large overcharge on cache hits (DeepSeek V4 hits are ~0.008-0.02x
+input, i.e. cached tokens cost 50-120x less than regular input).
 
 This module injects the missing entries into ``litellm.model_cost`` at startup
-so the existing backfill path resolves them. Rates mirror the open upstream PR
+so the existing backfill path resolves them. Rates mirror the canonical
+``deepseek`` provider entries now in litellm's ``model_prices`` map
+(``input_cost_per_token`` is the cache-*miss* rate;
+``cache_read_input_token_cost`` is the cache-*hit* rate), sourced from
+https://api-docs.deepseek.com/quick_start/pricing via
 https://github.com/BerriAI/litellm/pull/26380 (issue
 https://github.com/BerriAI/litellm/issues/30430).
 
@@ -22,21 +27,23 @@ from ..core import get_logger
 
 logger = get_logger(__name__)
 
-# USD per token. Source: BerriAI/litellm PR #26380.
+# USD per token. Mirrors the canonical ``deepseek`` provider entries in
+# litellm's model_prices map (source: DeepSeek API pricing docs). Keep these in
+# sync with ``litellm.model_cost["deepseek/deepseek-v4-*"]``.
 _DEEPSEEK_V4_RATES: dict[str, dict[str, float]] = {
     "deepseek-v4-flash": {
         "input_cost_per_token": 1.4e-07,
         "output_cost_per_token": 2.8e-07,
-        "cache_read_input_token_cost": 2.8e-08,
+        "cache_read_input_token_cost": 2.8e-09,
         "cache_creation_input_token_cost": 0.0,
-        "input_cost_per_token_cache_hit": 2.8e-08,
+        "input_cost_per_token_cache_hit": 2.8e-09,
     },
     "deepseek-v4-pro": {
-        "input_cost_per_token": 1.74e-06,
-        "output_cost_per_token": 3.48e-06,
-        "cache_read_input_token_cost": 1.4e-07,
+        "input_cost_per_token": 4.35e-07,
+        "output_cost_per_token": 8.7e-07,
+        "cache_read_input_token_cost": 3.625e-09,
         "cache_creation_input_token_cost": 0.0,
-        "input_cost_per_token_cache_hit": 1.4e-07,
+        "input_cost_per_token_cache_hit": 3.625e-09,
     },
 }
 

routstr/upstream/fireworks.py

@@ -20,7 +20,7 @@ class FireworksUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "FireworksUpstreamProvider":
         return cls(

routstr/upstream/gemini.py

@@ -51,7 +51,7 @@ class GeminiUpstreamProvider(BaseUpstreamProvider):
         return self._client
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "GeminiUpstreamProvider":
         return cls(

routstr/upstream/generic.py

@@ -5,6 +5,12 @@ from typing import TYPE_CHECKING
 import httpx
 
 from .base import BaseUpstreamProvider
+from .pricing_resolver import (
+    FallbackPricingResolver,
+    ResolvedPricing,
+    _as_float,
+    estimate_context_length,
+)
 
 if TYPE_CHECKING:
     from ..core.db import UpstreamProviderRow
@@ -45,7 +51,7 @@ class GenericUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "GenericUpstreamProvider":
         return cls(
@@ -64,6 +70,40 @@ class GenericUpstreamProvider(BaseUpstreamProvider):
             "platform_url": cls.platform_url,
         }
 
+    def _native_pricing(
+        self, model_id: str, model_spec: dict
+    ) -> ResolvedPricing | None:
+        """Read pricing/metadata from Venice's bespoke ``model_spec`` schema.
+
+        Returns ``None`` when the upstream reported no *usable* native price —
+        absent, non-numeric, negative, or both-zero — so the caller falls
+        through to the shared resolution chain instead of fabricating a number
+        or trusting a bogus one. This mirrors the money-safety guards the
+        litellm and OpenRouter rungs already apply: a both-zero price would
+        serve the model free, a negative one would credit the caller, and a
+        non-numeric string would otherwise throw and drop the whole catalog.
+        """
+        pricing_info = model_spec.get("pricing", {})
+        input_usd = _as_float(pricing_info.get("input", {}).get("usd"))
+        output_usd = _as_float(pricing_info.get("output", {}).get("usd"))
+        if input_usd is None or output_usd is None:
+            return None
+        if input_usd < 0 or output_usd < 0 or (input_usd == 0 and output_usd == 0):
+            return None
+
+        capabilities = model_spec.get("capabilities", {})
+        input_modalities = ["text"]
+        if capabilities.get("supportsVision", False):
+            input_modalities.append("image")
+
+        return ResolvedPricing(
+            prompt=input_usd / 1_000_000,
+            completion=output_usd / 1_000_000,
+            context_length=model_spec.get("availableContextTokens"),
+            source="native",
+            input_modalities=input_modalities,
+        )
+
     async def fetch_models(self) -> list[Model]:
         """Fetch models from upstream API using /models endpoint."""
         from ..payment.models import Architecture, Model, Pricing, TopProvider
@@ -78,6 +118,7 @@ class GenericUpstreamProvider(BaseUpstreamProvider):
                 response.raise_for_status()
                 data = response.json()
 
+                resolver = FallbackPricingResolver()
                 models_list = []
                 for model_data in data.get("data", []):
                     model_id = model_data.get("id", "")
@@ -89,41 +130,44 @@ class GenericUpstreamProvider(BaseUpstreamProvider):
                     owned_by = model_data.get("owned_by", "unknown")
                     model_spec = model_data.get("model_spec", {})
 
-                    context_length = 4096
-                    if model_spec.get("availableContextTokens"):
-                        context_length = model_spec["availableContextTokens"]
-                    elif any(
-                        pattern in model_id.lower() for pattern in ["32k", "32000"]
-                    ):
-                        context_length = 32768
-                    elif any(
-                        pattern in model_id.lower() for pattern in ["16k", "16000"]
-                    ):
-                        context_length = 16384
-                    elif any(pattern in model_id.lower() for pattern in ["8k", "8000"]):
-                        context_length = 8192
-                    elif "gpt-4" in model_id.lower():
-                        context_length = 8192
-                    elif "claude" in model_id.lower():
-                        context_length = 200000
+                    resolved = self._native_pricing(model_id, model_spec)
+                    if resolved is None:
+                        resolved = await resolver.resolve(model_id)
 
-                    pricing_info = model_spec.get("pricing", {})
-                    input_pricing = pricing_info.get("input", {})
-                    output_pricing = pricing_info.get("output", {})
+                    if resolved is None:
+                        # Fail closed: never invent a price. Import the model
+                        # disabled with a warning so the operator can price it
+                        # (the admin UI surfaces disabled remote models).
+                        logger.warning(
+                            f"No pricing source resolved for '{model_id}' from "
+                            f"{self.upstream_name}; importing it disabled",
+                            extra={"model_id": model_id, "base_url": self.base_url},
+                        )
+                        resolved = ResolvedPricing(
+                            prompt=0.0,
+                            completion=0.0,
+                            context_length=None,
+                            source="unresolved",
+                        )
+                        enabled = False
+                    else:
+                        enabled = True
 
-                    prompt_price = input_pricing.get("usd", 0.001) / 1000000
-                    completion_price = output_pricing.get("usd", 0.001) / 1000000
+                    # Prefer the source's own modality string (OpenRouter ships
+                    # one, e.g. "text+image->text"); otherwise derive it from the
+                    # captured input/output modalities in the same "in->out" shape
+                    # rather than flattening vision models to "text->text".
+                    modality = resolved.modality or (
+                        f"{'+'.join(resolved.input_modalities)}"
+                        f"->{'+'.join(resolved.output_modalities)}"
+                    )
 
-                    capabilities = model_spec.get("capabilities", {})
-                    input_modalities = ["text"]
-                    output_modalities = ["text"]
-
-                    if capabilities.get("supportsVision", False):
-                        input_modalities.append("image")
-
-                    modality = "text"
-                    if capabilities.get("supportsVision", False):
-                        modality = "text->text"
+                    # A source can carry a price but no context (e.g. a litellm
+                    # entry missing max_input_tokens); fall back to an id-based
+                    # estimate so we never persist a zero-length window.
+                    context_length = resolved.context_length or estimate_context_length(
+                        model_id
+                    )
 
                     spec_name = model_spec.get("name", model_name)
                     description = f"{spec_name}"
@@ -139,30 +183,33 @@ class GenericUpstreamProvider(BaseUpstreamProvider):
                             context_length=context_length,
                             architecture=Architecture(
                                 modality=modality,
-                                input_modalities=input_modalities,
-                                output_modalities=output_modalities,
-                                tokenizer="unknown",
-                                instruct_type=None,
+                                input_modalities=resolved.input_modalities,
+                                output_modalities=resolved.output_modalities,
+                                tokenizer=resolved.tokenizer,
+                                instruct_type=resolved.instruct_type,
                             ),
                             pricing=Pricing(
-                                prompt=prompt_price,
-                                completion=completion_price,
+                                prompt=resolved.prompt,
+                                completion=resolved.completion,
                                 request=0.0,
                                 image=0.0,
                                 web_search=0.0,
                                 internal_reasoning=0.0,
-                                max_prompt_cost=0.001,
-                                max_completion_cost=0.001,
-                                max_cost=0.001,
+                                input_cache_read=resolved.input_cache_read,
+                                input_cache_write=resolved.input_cache_write,
                             ),
                             sats_pricing=None,
                             per_request_limits=None,
                             top_provider=TopProvider(
                                 context_length=context_length,
-                                max_completion_tokens=context_length // 2,
-                                is_moderated=False,
+                                max_completion_tokens=(
+                                    resolved.max_completion_tokens
+                                    if resolved.max_completion_tokens is not None
+                                    else context_length // 2
+                                ),
+                                is_moderated=bool(resolved.is_moderated),
                             ),
-                            enabled=True,
+                            enabled=enabled,
                             upstream_provider_id=None,
                             canonical_slug=None,
                         )

routstr/upstream/groq.py

@@ -20,7 +20,7 @@ class GroqUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(cls, provider_row: "UpstreamProviderRow") -> "GroqUpstreamProvider":
+    def _build_from_row(cls, provider_row: "UpstreamProviderRow") -> "GroqUpstreamProvider":
         return cls(
             api_key=provider_row.api_key,
             provider_fee=provider_row.provider_fee,

routstr/upstream/helpers.py

@@ -12,6 +12,7 @@ from sqlmodel import select
 
 from ..core import get_logger
 from ..core.db import AsyncSession, ModelRow, UpstreamProviderRow, create_session
+from ..core.provider_slugs import allocate_unique_provider_slug
 from ..payment.models import Model
 from .base import BaseUpstreamProvider
 
@@ -215,9 +216,6 @@ async def init_upstreams() -> list[BaseUpstreamProvider]:
 
             provider = _instantiate_provider(provider_row)
             if provider:
-                # Keep provider DB id on runtime instance so model mapping can
-                # bind DB overrides to the correct upstream.
-                setattr(provider, "db_id", provider_row.id)
                 await provider.refresh_models_cache()
                 logger.debug(
                     f"Initialized {provider_row.provider_type} provider",
@@ -250,6 +248,7 @@ async def _seed_providers_from_settings(
 
     providers_to_add: list[UpstreamProviderRow] = []
     seeded_provider_keys: set[tuple[str, str]] = set()
+    reserved_slugs: set[str] = set()
 
     provider_classes_by_type = {
         cls.provider_type: cls
@@ -279,8 +278,13 @@ async def _seed_providers_from_settings(
                     )
                 )
                 if not result.first():
+                    slug = await allocate_unique_provider_slug(
+                        session, provider_type, reserved_slugs
+                    )
+                    reserved_slugs.add(slug)
                     providers_to_add.append(
                         UpstreamProviderRow(
+                            slug=slug,
                             provider_type=provider_type,
                             base_url=base_url,
                             api_key=api_key,
@@ -299,8 +303,13 @@ async def _seed_providers_from_settings(
             )
         )
         if not result.first():
+            slug = await allocate_unique_provider_slug(
+                session, "ollama", reserved_slugs
+            )
+            reserved_slugs.add(slug)
             providers_to_add.append(
                 UpstreamProviderRow(
+                    slug=slug,
                     provider_type="ollama",
                     base_url=ollama_base_url,
                     api_key=ollama_api_key,
@@ -320,8 +329,13 @@ async def _seed_providers_from_settings(
                 )
             )
             if not result.first():
+                slug = await allocate_unique_provider_slug(
+                    session, "azure", reserved_slugs
+                )
+                reserved_slugs.add(slug)
                 providers_to_add.append(
                     UpstreamProviderRow(
+                        slug=slug,
                         provider_type="azure",
                         base_url=base_url,
                         api_key=api_key,
@@ -342,8 +356,13 @@ async def _seed_providers_from_settings(
                 )
             )
             if not result.first():
+                slug = await allocate_unique_provider_slug(
+                    session, "custom", reserved_slugs
+                )
+                reserved_slugs.add(slug)
                 providers_to_add.append(
                     UpstreamProviderRow(
+                        slug=slug,
                         provider_type="custom",
                         base_url=base_url,
                         api_key=api_key,
@@ -356,7 +375,7 @@ async def _seed_providers_from_settings(
         session.add(provider)
         logger.info(
             f"Seeding {provider.provider_type} provider",  # type: ignore[str-format]
-            extra={"base_url": provider.base_url},
+            extra={"base_url": provider.base_url, "slug": provider.slug},
         )
 
 
@@ -391,9 +410,7 @@ def _instantiate_provider(
             return provider
 
         if provider_row.provider_type == "custom":
-            return BaseUpstreamProvider(
-                provider_row.base_url, provider_row.api_key, provider_row.provider_fee
-            )
+            return BaseUpstreamProvider.from_db_row(provider_row)
 
         logger.error(
             f"Unknown provider type: {provider_row.provider_type}",

routstr/upstream/messages_dispatch.py

@@ -29,7 +29,9 @@ import litellm
 
 from ..core import get_logger
 from ..core.exceptions import UpstreamError
+from ..core.redaction import redact_org_ids
 from ..payment.models import Model
+from .rate_limit import classify_rate_limit
 
 logger = get_logger(__name__)
 
@@ -505,23 +507,33 @@ async def dispatch_anthropic_messages(
     try:
         result = await litellm.anthropic.messages.acreate(**kwargs)
     except Exception as exc:
-        exc_message = getattr(exc, "message", None) or str(exc) or repr(exc)
+        raw_message = getattr(exc, "message", None) or str(exc) or repr(exc)
+        # Redact provider account identifiers before the message reaches logs
+        # or the surfaced error.
+        exc_message = redact_org_ids(raw_message)
         exc_status = getattr(exc, "status_code", None)
         exc_response = getattr(exc, "response", None)
         response_text = None
         if exc_response is not None:
             try:
-                response_text = getattr(exc_response, "text", str(exc_response))
+                response_text = redact_org_ids(
+                    getattr(exc_response, "text", str(exc_response))
+                )
             except Exception:
                 response_text = "<unreadable>"
+        status_for_classify = exc_status if isinstance(exc_status, int) else 502
+        rate_limit = classify_rate_limit(
+            status_for_classify, exc_message, getattr(exc, "headers", None)
+        )
         logger.error(
             "litellm dispatch failed",
             extra={
                 "error": exc_message,
                 "error_type": type(exc).__name__,
                 "status_code": exc_status,
+                "error_code": rate_limit.code if rate_limit else None,
                 "llm_provider": getattr(exc, "llm_provider", None),
-                "body": getattr(exc, "body", None),
+                "body": redact_org_ids(str(getattr(exc, "body", "") or "")) or None,
                 "response_text": response_text,
                 "model": litellm_model,
                 "api_base": base_url,
@@ -529,7 +541,9 @@ async def dispatch_anthropic_messages(
         )
         raise UpstreamError(
             f"Upstream error via litellm: {exc_message}",
-            status_code=exc_status if isinstance(exc_status, int) else 502,
+            status_code=status_for_classify,
+            code=rate_limit.code if rate_limit else None,
+            details=rate_limit.as_details() if rate_limit else None,
         ) from exc
 
     if not client_stream and hasattr(result, "__aiter__"):

routstr/upstream/ollama.py

@@ -43,7 +43,7 @@ class OllamaUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "OllamaUpstreamProvider":
         return cls(

routstr/upstream/openai.py

@@ -20,7 +20,7 @@ class OpenAIUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "OpenAIUpstreamProvider":
         return cls(

routstr/upstream/openrouter.py

@@ -90,7 +90,7 @@ class OpenRouterUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "OpenRouterUpstreamProvider":
         return cls(

routstr/upstream/perplexity.py

@@ -23,7 +23,7 @@ class PerplexityUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "PerplexityUpstreamProvider":
         return cls(

routstr/upstream/ppqai.py

@@ -46,7 +46,7 @@ class PPQAIUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "PPQAIUpstreamProvider":
         return cls(
@@ -229,17 +229,14 @@ class PPQAIUpstreamProvider(BaseUpstreamProvider):
                 f"Disabling PPQ.AI provider ({self.base_url}) due to insufficient balance",
                 extra={"error": error_message},
             )
-            from sqlmodel import select
-
             from ..core.db import UpstreamProviderRow, create_session
 
             async with create_session() as session:
-                statement = select(UpstreamProviderRow).where(
-                    UpstreamProviderRow.base_url == self.base_url,
-                    UpstreamProviderRow.api_key == self.api_key,
+                provider = (
+                    await session.get(UpstreamProviderRow, self.db_id)
+                    if self.db_id is not None
+                    else None
                 )
-                result = await session.exec(statement)
-                provider = result.first()
 
                 if provider:
                     provider.enabled = False

routstr/upstream/pricing_resolver.py

@@ -0,0 +1,203 @@
+"""Shared price/metadata resolution chain for upstream model discovery.
+
+Most OpenAI-compatible ``/models`` responses carry no pricing. Rather than let
+a provider fabricate one, this module resolves a model through decreasingly
+trustworthy sources — litellm's bundled cost map (curated list prices, mirrors
+provider docs), then the OpenRouter feed (resale prices, broader coverage) —
+and returns ``None`` when none of them know the model, so the caller can fail
+closed instead of inventing a number.
+
+Provider-native pricing (a gateway's own ``/models`` schema, e.g. Venice's
+``model_spec``) is authoritative and handled by the provider before this chain
+is consulted; only the shared fallback lives here so a later refactor can hoist
+it into the base provider unchanged.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+
+@dataclass
+class ResolvedPricing:
+    """Per-token pricing plus whatever metadata the answering source carried.
+
+    Prices are USD per token. ``source`` records provenance
+    (``native``/``litellm``/``openrouter``/``unresolved``) so later work can
+    surface where each price came from.
+    """
+
+    prompt: float
+    completion: float
+    context_length: int | None
+    source: str
+    modality: str | None = None
+    max_completion_tokens: int | None = None
+    input_cache_read: float = 0.0
+    input_cache_write: float = 0.0
+    input_modalities: list[str] = field(default_factory=lambda: ["text"])
+    output_modalities: list[str] = field(default_factory=lambda: ["text"])
+    tokenizer: str = "unknown"
+    instruct_type: str | None = None
+    is_moderated: bool | None = None
+
+
+def estimate_context_length(model_id: str) -> int:
+    """Best-effort context window from a model id when no source reports one.
+
+    The last rung of the fallback chain, reached only for a model whose price
+    resolved but whose context did not (or that imported disabled). Context is
+    not a billing input, so a rough id-based guess is acceptable here where a
+    guessed *price* never would be.
+    """
+    lowered = model_id.lower()
+    if any(pattern in lowered for pattern in ["32k", "32000"]):
+        return 32768
+    if any(pattern in lowered for pattern in ["16k", "16000"]):
+        return 16384
+    if any(pattern in lowered for pattern in ["8k", "8000"]):
+        return 8192
+    if "gpt-4" in lowered:
+        return 8192
+    if "claude" in lowered:
+        return 200000
+    return 4096
+
+
+def _as_float(value: object) -> float | None:
+    """OpenRouter reports prices as strings; coerce, ``None`` if unparseable."""
+    try:
+        return float(value)  # type: ignore[arg-type]
+    except (TypeError, ValueError):
+        return None
+
+
+def _as_int(value: object) -> int | None:
+    """Coerce an already-numeric token count to ``int``, else ``None``."""
+    return int(value) if isinstance(value, (int, float)) else None
+
+
+def _from_litellm(model_id: str) -> ResolvedPricing | None:
+    # Lazy import so the resolver stays import-light and shares the exact
+    # lookup semantics used by cache-rate backfill.
+    from ..payment.models import litellm_cost_entry
+
+    info = litellm_cost_entry(model_id)
+    if info is None:
+        return None
+
+    prompt = info.get("input_cost_per_token")
+    completion = info.get("output_cost_per_token")
+    if not isinstance(prompt, (int, float)) or not isinstance(completion, (int, float)):
+        return None
+    # A both-zero entry is litellm listing a model without a real price (free
+    # moderation/rerank tiers do this) — treating 0/0 as resolved would serve
+    # the model for free. Reject it (and any negative) so the caller falls
+    # through, mirroring async_fetch_openrouter_models' _has_valid_pricing.
+    if prompt < 0 or completion < 0 or (prompt == 0 and completion == 0):
+        return None
+
+    input_modalities = ["text"]
+    if info.get("supports_vision"):
+        input_modalities.append("image")
+
+    return ResolvedPricing(
+        prompt=float(prompt),
+        completion=float(completion),
+        # max_input_tokens is the context window; max_tokens is litellm's
+        # completion cap (it tracks max_output_tokens for ~94% of models), so
+        # it is never a context source. A missing window falls to the id-based
+        # estimate downstream rather than borrowing the output cap.
+        context_length=_as_int(info.get("max_input_tokens")),
+        source="litellm",
+        max_completion_tokens=_as_int(info.get("max_output_tokens")),
+        input_cache_read=float(info.get("cache_read_input_token_cost") or 0.0),
+        input_cache_write=float(info.get("cache_creation_input_token_cost") or 0.0),
+        input_modalities=input_modalities,
+    )
+
+
+def _match_openrouter(model_id: str, feed: list[dict]) -> dict | None:
+    """Find ``model_id`` in the OpenRouter feed, exact id before bare tail.
+
+    Bare-tail matching (``deepseek-chat`` ↔ ``deepseek/deepseek-chat``) is a
+    looser, lower-trust match — OpenRouter fans a model out across resellers —
+    so an exact id match always wins first. When several entries share the bare
+    tail, the one with the highest *combined* (prompt + completion) per-token
+    cost wins: the choice must be deterministic (not feed-order-dependent) and
+    money-safe whichever way traffic leans, since undercharging is the hazard.
+    Ranking on prompt alone could pick an entry that is cheap on input but dear
+    on output. The live feed has no such collisions today; this only governs
+    the latent case.
+    """
+    bare = model_id.split("/", 1)[-1]
+    exact = next((m for m in feed if m.get("id") == model_id), None)
+    if exact is not None:
+        return exact
+    matches = [m for m in feed if m.get("id", "").split("/", 1)[-1] == bare]
+    if not matches:
+        return None
+
+    def _combined_cost(m: dict) -> float:
+        pricing = m.get("pricing", {})
+        return (_as_float(pricing.get("prompt")) or 0.0) + (
+            _as_float(pricing.get("completion")) or 0.0
+        )
+
+    return max(matches, key=_combined_cost)
+
+
+def _from_openrouter(model_id: str, feed: list[dict]) -> ResolvedPricing | None:
+    entry = _match_openrouter(model_id, feed)
+    if entry is None:
+        return None
+
+    pricing = entry.get("pricing", {})
+    prompt = _as_float(pricing.get("prompt"))
+    completion = _as_float(pricing.get("completion"))
+    if prompt is None or completion is None:
+        return None
+
+    architecture = entry.get("architecture", {})
+    top_provider = entry.get("top_provider", {})
+
+    return ResolvedPricing(
+        prompt=prompt,
+        completion=completion,
+        context_length=_as_int(entry.get("context_length")),
+        source="openrouter",
+        modality=architecture.get("modality"),
+        max_completion_tokens=_as_int(top_provider.get("max_completion_tokens")),
+        input_cache_read=_as_float(pricing.get("input_cache_read")) or 0.0,
+        input_cache_write=_as_float(pricing.get("input_cache_write")) or 0.0,
+        input_modalities=architecture.get("input_modalities") or ["text"],
+        output_modalities=architecture.get("output_modalities") or ["text"],
+        tokenizer=architecture.get("tokenizer") or "unknown",
+        instruct_type=architecture.get("instruct_type"),
+        is_moderated=top_provider.get("is_moderated"),
+    )
+
+
+class FallbackPricingResolver:
+    """Resolves models via litellm → OpenRouter for one discovery pass.
+
+    The OpenRouter catalog is fetched at most once and only when a model
+    actually misses litellm, so a provider full of litellm-known models never
+    touches the network. Instantiate one per ``fetch_models`` call.
+    """
+
+    def __init__(self) -> None:
+        self._openrouter_feed: list[dict] | None = None
+
+    async def resolve(self, model_id: str) -> ResolvedPricing | None:
+        """Resolve ``model_id``; ``None`` if no source knows it."""
+        resolved = _from_litellm(model_id)
+        if resolved is not None:
+            return resolved
+
+        if self._openrouter_feed is None:
+            # Lazy import so tests can patch the feed at its source.
+            from ..payment.models import async_fetch_openrouter_models
+
+            self._openrouter_feed = await async_fetch_openrouter_models()
+        return _from_openrouter(model_id, self._openrouter_feed)

routstr/upstream/rate_limit.py

@@ -0,0 +1,136 @@
+"""Detection and parsing of upstream provider rate-limit errors.
+
+Upstream OpenAI-compatible providers signal rate limits via HTTP 429 and/or a
+human-readable message such as::
+
+    Rate limit reached for gpt-5.5-2026-04-23 (for limit gpt-5.5) in organization
+    org-XXXX on tokens per min (TPM): Limit 180000000, Used 180000000,
+    Requested 8929. Please try again in 2ms.
+
+This module classifies those failures into a stable :data:`UPSTREAM_RATE_LIMIT`
+code and extracts useful debugging fields. All retained text is redacted of
+organization IDs first.
+"""
+
+from __future__ import annotations
+
+import re
+from dataclasses import asdict, dataclass
+
+from ..core.redaction import redact_org_ids
+
+# Stable error code callers can switch on to distinguish upstream rate limits
+# from generic request failures. The literal value matches the identifier named
+# in issue #555 ("UPSTREAM_RATE_LIMIT") so the public API contract is exact.
+UPSTREAM_RATE_LIMIT = "UPSTREAM_RATE_LIMIT"
+
+# Message fragments that indicate a rate-limit even when the status code is not
+# 429 (some providers wrap it in a 400/500 envelope).
+_RATE_LIMIT_MARKERS = (
+    "rate limit reached",
+    "rate_limit_exceeded",
+    "rate limit exceeded",
+    "too many requests",
+)
+
+_MODEL_RE = re.compile(r"Rate limit reached for ([^\s(]+)", re.IGNORECASE)
+_LIMIT_NAME_RE = re.compile(r"\(for limit ([^)]+)\)", re.IGNORECASE)
+_METRIC_RE = re.compile(r"on ([a-z ]+\((?:TPM|RPM|TPD|RPD|IPM)\))", re.IGNORECASE)
+_LIMIT_RE = re.compile(r"Limit (\d+)", re.IGNORECASE)
+_USED_RE = re.compile(r"Used (\d+)", re.IGNORECASE)
+_REQUESTED_RE = re.compile(r"Requested (\d+)", re.IGNORECASE)
+_RETRY_RE = re.compile(r"try again in ([\d.]+)\s*(ms|s)", re.IGNORECASE)
+
+
+@dataclass
+class RateLimitInfo:
+    """Structured, redaction-safe view of an upstream rate-limit error."""
+
+    code: str
+    message: str
+    model: str | None = None
+    limit_name: str | None = None
+    metric: str | None = None
+    limit: int | None = None
+    used: int | None = None
+    requested: int | None = None
+    retry_after_seconds: float | None = None
+
+    def as_details(self) -> dict[str, object]:
+        """Return a JSON-serialisable dict for embedding in an error envelope."""
+        return {k: v for k, v in asdict(self).items() if v is not None}
+
+
+def _looks_like_rate_limit(status_code: int, message: str) -> bool:
+    if status_code == 429:
+        return True
+    lowered = message.lower()
+    return any(marker in lowered for marker in _RATE_LIMIT_MARKERS)
+
+
+def _parse_retry_after_header(headers: dict[str, str] | None) -> float | None:
+    """Parse a ``Retry-After`` header (delta-seconds form) into seconds."""
+    if not headers:
+        return None
+    raw = headers.get("retry-after") or headers.get("Retry-After")
+    if raw is None:
+        return None
+    try:
+        return float(str(raw).strip())
+    except (TypeError, ValueError):
+        return None
+
+
+def _int_or_none(match: re.Match[str] | None) -> int | None:
+    if match is None:
+        return None
+    try:
+        return int(match.group(1))
+    except (TypeError, ValueError):
+        return None
+
+
+def classify_rate_limit(
+    status_code: int,
+    message: str,
+    headers: dict[str, str] | None = None,
+) -> RateLimitInfo | None:
+    """Classify an upstream error as a rate-limit and extract its fields.
+
+    Args:
+        status_code: HTTP status code from the upstream response.
+        message: Upstream error message (may contain sensitive identifiers).
+        headers: Optional upstream response headers, used for ``Retry-After``.
+
+    Returns:
+        A :class:`RateLimitInfo` when the error is a rate-limit, else ``None``.
+    """
+    message = message or ""
+    if not _looks_like_rate_limit(status_code, message):
+        return None
+
+    redacted = redact_org_ids(message)
+
+    model_match = _MODEL_RE.search(redacted)
+    metric_match = _METRIC_RE.search(redacted)
+
+    retry_after = _parse_retry_after_header(headers)
+    if retry_after is None:
+        retry_match = _RETRY_RE.search(redacted)
+        if retry_match is not None:
+            value = float(retry_match.group(1))
+            retry_after = value / 1000.0 if retry_match.group(2).lower() == "ms" else value
+
+    limit_name_match = _LIMIT_NAME_RE.search(redacted)
+
+    return RateLimitInfo(
+        code=UPSTREAM_RATE_LIMIT,
+        message=redacted,
+        model=model_match.group(1) if model_match else None,
+        limit_name=limit_name_match.group(1).strip() if limit_name_match else None,
+        metric=metric_match.group(1).strip() if metric_match else None,
+        limit=_int_or_none(_LIMIT_RE.search(redacted)),
+        used=_int_or_none(_USED_RE.search(redacted)),
+        requested=_int_or_none(_REQUESTED_RE.search(redacted)),
+        retry_after_seconds=retry_after,
+    )

routstr/upstream/routstr.py

@@ -55,7 +55,7 @@ class RoutstrUpstreamProvider(BaseUpstreamProvider):
         return path.lstrip("/")
 
     @classmethod
-    def from_db_row(
+    def _build_from_row(
         cls, provider_row: "UpstreamProviderRow"
     ) -> "RoutstrUpstreamProvider":
         import json

routstr/upstream/xai.py

@@ -21,7 +21,7 @@ class XAIUpstreamProvider(BaseUpstreamProvider):
         )
 
     @classmethod
-    def from_db_row(cls, provider_row: "UpstreamProviderRow") -> "XAIUpstreamProvider":
+    def _build_from_row(cls, provider_row: "UpstreamProviderRow") -> "XAIUpstreamProvider":
         return cls(
             api_key=provider_row.api_key,
             provider_fee=provider_row.provider_fee,

routstr/wallet.py

@@ -1,4 +1,5 @@
 import asyncio
+import re
 import time
 import typing
 from typing import TypedDict
@@ -129,6 +130,72 @@ async def send_token(amount: int, unit: str, mint_url: str | None = None) -> str
     return token
 
 
+# A foreign mint's fee_reserve is a non-binding estimate (NUT-05): the mint may
+# demand more when re-quoting or at melt execution. Instead of padding the
+# estimate with a safety buffer (which strands the margin at the foreign mint
+# on every swap), the swap retries with the amount recomputed from the fees the
+# mint actually demands, up to this many attempts.
+_MAX_SWAP_ATTEMPTS = 3
+
+_MINT_ERROR_CODE_RE = re.compile(r"\(Code: (\d+)\)")
+_MELT_SHORTFALL_RE = re.compile(r"Provided: (\d+), needed: (\d+)")
+
+# Insufficient-melt-inputs failures differ across mint implementations. 11005 is
+# the registered "Transaction is not balanced" code (cdk), specific enough to
+# trust on the code alone. 11000 is nutshell's generic, unregistered
+# TransactionError covering many unrelated failures, so it only counts as a fee
+# shortfall alongside the "not enough inputs" detail text. With no code suffix at
+# all, that same text is the only signal.
+
+
+def _net_minted_amount(amount_msat: int, token_unit: str, fees: int) -> int:
+    """
+    Convert the token value minus fees (given in the token unit) into an
+    amount in the primary mint's unit.
+    """
+    fee_msat = fees * 1000 if token_unit == "sat" else fees
+    remaining_msat = amount_msat - fee_msat
+    if settings.primary_mint_unit == "sat":
+        return int(remaining_msat // 1000)
+    return int(remaining_msat)
+
+
+def _melt_insufficient_shortfall(error: Exception) -> int | None:
+    """
+    Classify a melt failure: return the observed shortfall (in the token unit)
+    when the mint rejected the inputs as insufficient, or None when the failure
+    is unrelated to fees and must not be retried (e.g. a Lightning payment
+    failure, where a smaller invoice would not help).
+
+    Cashu errors carry no structured amounts (NUT-00 defines only detail/code,
+    flattened to "Mint Error: <detail> (Code: <code>)" by cashu-py), so the
+    classification uses the code and the shortfall must be inferred: the
+    "Provided: X, needed: Y" amounts are nutshell-specific free text and only
+    refine the shortfall when present; otherwise shrink one unit at a time.
+    """
+    message = str(error)
+    code_match = _MINT_ERROR_CODE_RE.search(message)
+    code = code_match.group(1) if code_match is not None else None
+    has_shortfall_text = "not enough inputs" in message.lower()
+
+    match code:
+        case "11005":  # registered TransactionUnbalanced: trust the code
+            pass
+        case "11000" if has_shortfall_text:  # generic nutshell error: needs the text
+            pass
+        case None if has_shortfall_text:  # no code suffix: text is the only signal
+            pass
+        case _:  # other codes, a bare 11000, or no signal: must not retry
+            return None
+
+    amounts = _MELT_SHORTFALL_RE.search(message)
+    if amounts is not None:
+        provided, needed = int(amounts.group(1)), int(amounts.group(2))
+        if needed > provided:
+            return needed - provided
+    return 1
+
+
 async def _calculate_swap_amount(
     amount_msat: int,
     token_unit: str,
@@ -167,26 +234,18 @@ async def _calculate_swap_amount(
 
         fee_reserve = dummy_melt_quote.fee_reserve
         input_fees = token_wallet.get_fees_for_proofs(proofs)
-        if token_unit == "sat":
-            fee_msat = (fee_reserve + input_fees) * 1000
-        else:
-            fee_msat = fee_reserve + input_fees
-
-        amount_msat_after_fee = amount_msat - fee_msat
-
-        if settings.primary_mint_unit == "sat":
-            minted_amount = int(amount_msat_after_fee // 1000)
-        else:
-            minted_amount = int(amount_msat_after_fee)
+        total_fees = fee_reserve + input_fees
+        minted_amount = _net_minted_amount(amount_msat, token_unit, total_fees)
 
         if minted_amount <= 0:
-            raise ValueError(f"Fees ({fee_reserve + input_fees} {token_unit}) exceed token amount")
+            raise ValueError(f"Fees ({total_fees} {token_unit}) exceed token amount")
 
         logger.info(
             "swap_to_primary_mint: fee estimation result",
             extra={
                 "token_amount_sat": amount_msat // 1000,
-                "estimated_fee_sat": fee_msat // 1000,
+                "estimated_fee": total_fees,
+                "estimated_fee_unit": token_unit,
                 "input_fees": input_fees,
                 "minted_amount": minted_amount,
                 "minted_unit": settings.primary_mint_unit,
@@ -251,67 +310,116 @@ async def swap_to_primary_mint(
         token_obj.proofs,
     )
 
-    mint_quote = await primary_wallet.request_mint(minted_amount)
-    logger.info(
-        "swap_to_primary_mint: mint quote received",
-        extra={"mint_quote_id": mint_quote.quote},
-    )
+    # The estimate above is non-binding: the mint may demand a higher fee on the
+    # real quote or reject the melt outright. Retry the quote/melt cycle with the
+    # amount recomputed from the fees the mint actually demands.
+    observed_extra_fee = 0
+    attempt = 0
+    while True:
+        attempt += 1
+        mint_quote = await primary_wallet.request_mint(minted_amount)
+        logger.info(
+            "swap_to_primary_mint: mint quote received",
+            extra={"mint_quote_id": mint_quote.quote, "attempt": attempt},
+        )
 
-    melt_quote = await token_wallet.melt_quote(mint_quote.request)
-    input_fees = token_wallet.get_fees_for_proofs(token_obj.proofs)
-    total_needed = melt_quote.amount + melt_quote.fee_reserve + input_fees
-    logger.info(
-        "swap_to_primary_mint: melt quote received",
-        extra={
-            "melt_quote_id": melt_quote.quote,
-            "melt_amount": melt_quote.amount,
-            "melt_fee_reserve": melt_quote.fee_reserve,
-            "input_fees": input_fees,
-            "total_needed": total_needed,
-            "token_amount": token_amount,
-        },
-    )
-
-    if total_needed > token_amount:
-        logger.warning(
-            "swap_to_primary_mint: insufficient token amount for melt fees",
+        melt_quote = await token_wallet.melt_quote(mint_quote.request)
+        input_fees = token_wallet.get_fees_for_proofs(token_obj.proofs)
+        total_needed = melt_quote.amount + melt_quote.fee_reserve + input_fees
+        logger.info(
+            "swap_to_primary_mint: melt quote received",
             extra={
-                "token_amount": token_amount,
+                "melt_quote_id": melt_quote.quote,
                 "melt_amount": melt_quote.amount,
                 "melt_fee_reserve": melt_quote.fee_reserve,
                 "input_fees": input_fees,
                 "total_needed": total_needed,
-                "shortfall": total_needed - token_amount,
+                "token_amount": token_amount,
+                "attempt": attempt,
             },
         )
-        raise ValueError(
-            f"Token amount ({token_amount} {token_obj.unit}) is insufficient to cover "
-            f"melt fees. Needed: {total_needed} {token_obj.unit} "
-            f"(amount: {melt_quote.amount} + fee: {melt_quote.fee_reserve} + input_fees: {input_fees})"
-        )
 
-    try:
-        _ = await token_wallet.melt(
-            proofs=token_obj.proofs,
-            invoice=mint_quote.request,
-            fee_reserve_sat=melt_quote.fee_reserve,
-            quote_id=melt_quote.quote,
-        )
-    except Exception as e:
-        logger.error(
-            "swap_to_primary_mint: melt failed",
-            extra={
-                "error": str(e),
-                "error_type": type(e).__name__,
-                "foreign_mint": token_obj.mint,
-                "token_amount": token_amount,
-                "melt_quote_id": melt_quote.quote,
-                "total_needed": total_needed,
-            },
-        )
-        raise ValueError(
-            f"Failed to melt token from foreign mint {token_obj.mint}: {e}"
-        ) from e
+        if total_needed > token_amount:
+            recomputed = _net_minted_amount(
+                amount_msat,
+                token_obj.unit,
+                melt_quote.fee_reserve + input_fees + observed_extra_fee,
+            )
+            if attempt >= _MAX_SWAP_ATTEMPTS or recomputed <= 0:
+                logger.warning(
+                    "swap_to_primary_mint: insufficient token amount for melt fees",
+                    extra={
+                        "token_amount": token_amount,
+                        "melt_amount": melt_quote.amount,
+                        "melt_fee_reserve": melt_quote.fee_reserve,
+                        "input_fees": input_fees,
+                        "total_needed": total_needed,
+                        "shortfall": total_needed - token_amount,
+                        "attempts": attempt,
+                    },
+                )
+                raise ValueError(
+                    f"Token amount ({token_amount} {token_obj.unit}) is insufficient to cover "
+                    f"melt fees. Needed: {total_needed} {token_obj.unit} "
+                    f"(amount: {melt_quote.amount} + fee: {melt_quote.fee_reserve} + input_fees: {input_fees})"
+                )
+            logger.warning(
+                "swap_to_primary_mint: melt quote exceeds token amount, retrying",
+                extra={
+                    "total_needed": total_needed,
+                    "token_amount": token_amount,
+                    "retry_minted_amount": recomputed,
+                    "attempt": attempt,
+                },
+            )
+            minted_amount = recomputed
+            continue
+
+        try:
+            _ = await token_wallet.melt(
+                proofs=token_obj.proofs,
+                invoice=mint_quote.request,
+                fee_reserve_sat=melt_quote.fee_reserve,
+                quote_id=melt_quote.quote,
+            )
+        except Exception as e:
+            shortfall = _melt_insufficient_shortfall(e)
+            recomputed = 0
+            if shortfall is not None:
+                observed_extra_fee += shortfall
+                recomputed = _net_minted_amount(
+                    amount_msat,
+                    token_obj.unit,
+                    melt_quote.fee_reserve + input_fees + observed_extra_fee,
+                )
+            if shortfall is None or attempt >= _MAX_SWAP_ATTEMPTS or recomputed <= 0:
+                logger.error(
+                    "swap_to_primary_mint: melt failed",
+                    extra={
+                        "error": str(e),
+                        "error_type": type(e).__name__,
+                        "foreign_mint": token_obj.mint,
+                        "token_amount": token_amount,
+                        "melt_quote_id": melt_quote.quote,
+                        "total_needed": total_needed,
+                        "attempts": attempt,
+                    },
+                )
+                raise ValueError(
+                    f"Failed to melt token from foreign mint {token_obj.mint}: {e}"
+                ) from e
+            logger.warning(
+                "swap_to_primary_mint: mint demanded more than quoted at melt, retrying",
+                extra={
+                    "shortfall": shortfall,
+                    "retry_minted_amount": recomputed,
+                    "attempt": attempt,
+                },
+            )
+            minted_amount = recomputed
+            continue
+
+        break
 
     logger.info(
         "swap_to_primary_mint: melt succeeded, minting on primary",
@@ -441,7 +549,11 @@ async def credit_balance(
             .where(col(db.ApiKey.hashed_key) == key.hashed_key)
             .values(balance=(db.ApiKey.balance) + amount)
         )
-        await session.exec(stmt)  # type: ignore[call-overload]
+        result = await session.exec(stmt)  # type: ignore[call-overload]
+        # If pruning removed this key after redemption, do not commit a no-op
+        # balance update and pretend the top-up succeeded.
+        if (getattr(result, "rowcount", 0) or 0) == 0:
+            raise ValueError("API key disappeared before credit could be recorded")
         await session.commit()
         await session.refresh(key)
 

tests/integration/test_admin_model_cache_pricing.py

@@ -0,0 +1,229 @@
+import json
+from datetime import datetime, timedelta, timezone
+from unittest.mock import patch
+
+import pytest
+from httpx import AsyncClient
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.core.admin import admin_sessions
+from routstr.core.db import ModelRow, UpstreamProviderRow
+from routstr.payment.cost_calculation import CostData, calculate_cost
+from routstr.proxy import get_model_instance, reinitialize_upstreams
+
+
+def _admin_headers() -> dict[str, str]:
+    token = "test-admin-cache-pricing-token"
+    admin_sessions[token] = int(
+        (datetime.now(timezone.utc) + timedelta(minutes=5)).timestamp()
+    )
+    return {"Authorization": f"Bearer {token}"}
+
+
+def _model_payload(
+    provider_id: int,
+    *,
+    cache_read: float,
+    cache_write: float,
+) -> dict[str, object]:
+    return {
+        "id": "custom-cache-model",
+        "name": "Custom Cache Model",
+        "description": "custom model with explicit cache pricing",
+        "created": 0,
+        "context_length": 128000,
+        "architecture": {
+            "modality": "text",
+            "input_modalities": ["text"],
+            "output_modalities": ["text"],
+            "tokenizer": "unknown",
+            "instruct_type": None,
+        },
+        "pricing": {
+            "prompt": 1.4e-7,
+            "completion": 2.8e-7,
+            "input_cache_read": cache_read,
+            "input_cache_write": cache_write,
+            "request": 0.0,
+            "image": 0.0,
+            "web_search": 0.0,
+            "internal_reasoning": 0.0,
+        },
+        "per_request_limits": None,
+        "top_provider": None,
+        "upstream_provider_id": provider_id,
+        "canonical_slug": None,
+        "alias_ids": [],
+        "enabled": True,
+        "forwarded_model_id": "custom-cache-model",
+    }
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_admin_provider_model_api_persists_cache_pricing_on_create_and_update(
+    integration_client: AsyncClient,
+    integration_session: AsyncSession,
+) -> None:
+    provider = UpstreamProviderRow(
+        provider_type="generic",
+        base_url="https://custom-upstream.example/v1",
+        api_key="test-key",
+        provider_fee=1.0,
+    )
+    integration_session.add(provider)
+    await integration_session.commit()
+    await integration_session.refresh(provider)
+    assert provider.id is not None
+    await reinitialize_upstreams()
+
+    headers = _admin_headers()
+    create_payload = _model_payload(
+        provider.id,
+        cache_read=2.8e-9,
+        cache_write=3.5e-9,
+    )
+
+    with patch("routstr.payment.models.sats_usd_price", return_value=1e-6):
+        create_response = await integration_client.post(
+            f"/admin/api/upstream-providers/{provider.id}/models",
+            headers=headers,
+            json=create_payload,
+        )
+
+    assert create_response.status_code == 200
+    create_body = create_response.json()
+    assert create_body["pricing"]["input_cache_read"] == pytest.approx(2.8e-9)
+    assert create_body["pricing"]["input_cache_write"] == pytest.approx(3.5e-9)
+
+    row = await integration_session.get(ModelRow, ("custom-cache-model", provider.id))
+    assert row is not None
+    stored_pricing = json.loads(row.pricing)
+    assert stored_pricing["input_cache_read"] == pytest.approx(2.8e-9)
+    assert stored_pricing["input_cache_write"] == pytest.approx(3.5e-9)
+
+    update_payload = _model_payload(
+        provider.id,
+        cache_read=1.25e-9,
+        cache_write=4.5e-9,
+    )
+
+    with patch("routstr.payment.models.sats_usd_price", return_value=1e-6):
+        update_response = await integration_client.post(
+            f"/admin/api/upstream-providers/{provider.id}/models",
+            headers=headers,
+            json=update_payload,
+        )
+
+    assert update_response.status_code == 200
+    update_body = update_response.json()
+    assert update_body["pricing"]["input_cache_read"] == pytest.approx(1.25e-9)
+    assert update_body["pricing"]["input_cache_write"] == pytest.approx(4.5e-9)
+
+    await integration_session.refresh(row)
+    updated_pricing = json.loads(row.pricing)
+    assert updated_pricing["input_cache_read"] == pytest.approx(1.25e-9)
+    assert updated_pricing["input_cache_write"] == pytest.approx(4.5e-9)
+
+    model = get_model_instance("custom-cache-model")
+    assert model is not None
+    assert model.sats_pricing is not None
+    assert model.sats_pricing.input_cache_read == pytest.approx(0.00125)
+    assert model.sats_pricing.input_cache_write == pytest.approx(0.0045)
+
+    with patch("routstr.payment.cost_calculation.sats_usd_price", return_value=1e-6):
+        cost = await calculate_cost(
+            {
+                "model": "custom-cache-model",
+                "usage": {
+                    "prompt_tokens": 1000,
+                    "completion_tokens": 100,
+                    "prompt_tokens_details": {"cached_tokens": 800},
+                },
+            },
+            max_cost=1_000_000,
+        )
+
+    assert isinstance(cost, CostData)
+    assert cost.input_tokens == 200
+    assert cost.cache_read_input_tokens == 800
+    assert cost.cache_read_msats == 1000
+    assert cost.output_msats == 28000
+    assert cost.input_msats == 29000
+    assert cost.total_msats == 57000
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_upstream_response_cost_uses_model_cache_pricing(
+    integration_session: AsyncSession,
+    patched_db_engine: None,
+) -> None:
+    """A model's configured cache price must discount upstream cached-token usage."""
+    provider = UpstreamProviderRow(
+        provider_type="generic",
+        base_url="https://cache-priced-upstream.example/v1",
+        api_key="test-key",
+        provider_fee=1.0,
+    )
+    integration_session.add(provider)
+    await integration_session.commit()
+    await integration_session.refresh(provider)
+    assert provider.id is not None
+
+    row = ModelRow(
+        id="cache-priced-model",
+        name="Cache Priced Model",
+        description="model seeded with explicit cache pricing",
+        created=0,
+        context_length=128000,
+        architecture=json.dumps(
+            {
+                "modality": "text",
+                "input_modalities": ["text"],
+                "output_modalities": ["text"],
+                "tokenizer": "unknown",
+                "instruct_type": None,
+            }
+        ),
+        pricing=json.dumps(
+            {
+                "prompt": 1.4e-7,
+                "completion": 2.8e-7,
+                "input_cache_read": 1.25e-9,
+                "input_cache_write": 4.5e-9,
+            }
+        ),
+        upstream_provider_id=provider.id,
+        enabled=True,
+        forwarded_model_id="cache-priced-model",
+    )
+    integration_session.add(row)
+    await integration_session.commit()
+
+    with patch("routstr.payment.models.sats_usd_price", return_value=1e-6):
+        await reinitialize_upstreams()
+
+    with patch("routstr.payment.cost_calculation.sats_usd_price", return_value=1e-6):
+        cost = await calculate_cost(
+            {
+                "model": "cache-priced-model",
+                "usage": {
+                    "prompt_tokens": 1000,
+                    "completion_tokens": 100,
+                    "prompt_tokens_details": {"cached_tokens": 800},
+                },
+            },
+            max_cost=1_000_000,
+        )
+
+    assert isinstance(cost, CostData)
+    # prompt 1.4e-7 USD/token -> 0.14 sats/token -> 140 msats/token
+    # cache 1.25e-9 USD/token -> 0.00125 sats/token -> 1.25 msats/token
+    # completion 2.8e-7 USD/token -> 0.28 sats/token -> 280 msats/token
+    assert cost.input_tokens == 200
+    assert cost.cache_read_input_tokens == 800
+    assert cost.cache_read_msats == 1000
+    assert cost.input_msats == 29000
+    assert cost.output_msats == 28000
+    assert cost.total_msats == 57000

tests/integration/test_free_response_stale_reservation.py

@@ -0,0 +1,146 @@
+"""Regression tests for charging after stale reservation cleanup."""
+
+import time
+import uuid
+from unittest.mock import patch
+
+import pytest
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.core.db import ApiKey
+from routstr.payment.cost_calculation import CostData
+
+
+def _make_key(balance: int, reserved: int) -> ApiKey:
+    return ApiKey(
+        hashed_key=f"test_{uuid.uuid4().hex}",
+        balance=balance,
+        reserved_balance=reserved,
+        total_spent=0,
+        total_requests=1,
+    )
+
+
+def _cost_data(total_msats: int) -> CostData:
+    return CostData(
+        base_msats=0,
+        input_msats=total_msats // 2,
+        output_msats=total_msats - total_msats // 2,
+        total_msats=total_msats,
+        total_usd=0.0,
+        input_tokens=100,
+        output_tokens=100,
+    )
+
+
+@pytest.mark.asyncio
+async def test_overrun_charges_after_reservation_swept(
+    integration_session: AsyncSession,
+) -> None:
+    """Overrun finalize must charge even when the reservation was already released."""
+    from routstr.auth import adjust_payment_for_tokens
+
+    deducted_max_cost = 990  # discounted reservation
+    actual_token_cost = 1000  # actual cost overruns the reservation
+
+    # Sweeper has zeroed reserved_balance but left balance untouched.
+    key = _make_key(balance=1000, reserved=0)
+    integration_session.add(key)
+    await integration_session.commit()
+
+    response_data = {
+        "model": "test-model",
+        "usage": {"prompt_tokens": 100, "completion_tokens": 100},
+    }
+
+    with patch(
+        "routstr.auth.calculate_cost",
+        return_value=_cost_data(actual_token_cost),
+    ):
+        await adjust_payment_for_tokens(
+            key, response_data, integration_session, deducted_max_cost
+        )
+
+    await integration_session.refresh(key)
+
+    assert key.total_spent == actual_token_cost, (
+        f"Request was not billed (total_spent={key.total_spent}) — free response bug"
+    )
+    assert key.balance == 1000 - actual_token_cost, (
+        f"Balance not charged: {key.balance}"
+    )
+    assert key.balance >= 0
+    assert key.reserved_balance == 0
+
+
+@pytest.mark.asyncio
+async def test_free_response_path_closed_end_to_end(
+    integration_session: AsyncSession,
+    patched_db_engine: None,
+) -> None:
+    """A reservation released by the real sweeper must not yield a free response."""
+    from routstr.auth import adjust_payment_for_tokens, pay_for_request
+    from routstr.core.db import create_session, release_stale_reservations
+
+    deducted_max_cost = 990
+    actual_token_cost = 1000
+    key_hash = f"test_sweep_{uuid.uuid4().hex}"
+
+    async with create_session() as session:
+        session.add(
+            ApiKey(
+                hashed_key=key_hash,
+                balance=1000,
+                reserved_balance=0,
+                total_spent=0,
+                total_requests=0,
+            )
+        )
+        await session.commit()
+
+    # Reserve the request, then backdate reserved_at so the sweeper treats it as
+    # stale (simulates a stream that outlived stale_reservation_timeout_seconds).
+    async with create_session() as session:
+        key = await session.get(ApiKey, key_hash)
+        assert key is not None
+        await pay_for_request(key, deducted_max_cost, session)
+        await session.refresh(key)
+        assert key.reserved_balance == deducted_max_cost
+        key.reserved_at = int(time.time()) - 10_000
+        session.add(key)
+        await session.commit()
+
+    # Sweeper releases the stale reservation without charging.
+    async with create_session() as session:
+        released = await release_stale_reservations(session, max_age_seconds=300)
+        assert released == 1
+
+    async with create_session() as session:
+        key = await session.get(ApiKey, key_hash)
+        assert key is not None
+        assert key.reserved_balance == 0, "Precondition: sweeper zeroed the reservation"
+
+        response_data = {
+            "model": "test-model",
+            "usage": {"prompt_tokens": 100, "completion_tokens": 100},
+        }
+        with patch(
+            "routstr.auth.calculate_cost",
+            return_value=_cost_data(actual_token_cost),
+        ):
+            await adjust_payment_for_tokens(
+                key, response_data, session, deducted_max_cost
+            )
+
+    async with create_session() as session:
+        final = await session.get(ApiKey, key_hash)
+        assert final is not None
+
+    assert final.total_spent == actual_token_cost, (
+        f"Free response: total_spent={final.total_spent}, expected {actual_token_cost}"
+    )
+    assert final.balance == 1000 - actual_token_cost, (
+        f"Balance not charged after sweep: {final.balance}"
+    )
+    assert final.balance >= 0
+    assert final.reserved_balance == 0

tests/integration/test_provider_self_lookup.py

@@ -0,0 +1,128 @@
+"""A live upstream provider resolves its OWN database row by stable identity
+(its primary key), not by its mutable/secret ``api_key``.
+
+Today ``from_db_row`` drops ``provider_row.id`` and the two self-referential
+paths — PPQ.AI's insufficient-balance self-disable and the base
+``refresh_models_cache`` — re-find their own row with
+``WHERE base_url == self.base_url AND api_key == self.api_key``. That uses a
+rotatable secret as a self-handle: the moment the row's key changes underneath a
+live object (a rotation racing an in-flight request), the object can no longer
+find itself. These tests pin the invariant that a provider looks itself up by
+identity, so the lookup survives a key change (and, later, key encryption).
+"""
+
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from routstr.core.db import UpstreamProviderRow
+from routstr.upstream.ppqai import PPQAIUpstreamProvider
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_provider_object_carries_its_persistent_identity(
+    integration_session: object,
+    patched_db_engine: None,
+) -> None:
+    """``from_db_row`` gives the in-memory object its row's identity (``db_id``)."""
+    row = UpstreamProviderRow(
+        provider_type="ppqai",
+        base_url="https://api.ppq.ai",
+        api_key="sk-original",
+        enabled=True,
+        provider_fee=1.0,
+    )
+    integration_session.add(row)  # type: ignore[attr-defined]
+    await integration_session.commit()  # type: ignore[attr-defined]
+    await integration_session.refresh(row)  # type: ignore[attr-defined]
+
+    provider = PPQAIUpstreamProvider.from_db_row(row)
+    assert provider is not None
+
+    assert provider.db_id == row.id
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_self_disable_targets_own_row_after_key_rotation(
+    integration_session: object,
+    patched_db_engine: None,
+) -> None:
+    """PPQ.AI self-disable must disable *its* row even after the key rotated.
+
+    RED (current): the object holds the pre-rotation key, so the
+    ``(base_url, api_key)`` lookup misses the row → the provider is never
+    disabled. GREEN: lookup by ``id`` finds it and disables it.
+    """
+    row = UpstreamProviderRow(
+        provider_type="ppqai",
+        base_url="https://api.ppq.ai",
+        api_key="sk-original",
+        enabled=True,
+        provider_fee=1.0,
+    )
+    integration_session.add(row)  # type: ignore[attr-defined]
+    await integration_session.commit()  # type: ignore[attr-defined]
+    await integration_session.refresh(row)  # type: ignore[attr-defined]
+
+    provider = PPQAIUpstreamProvider.from_db_row(row)  # captures sk-original
+    assert provider is not None
+
+    # Key is rotated in the DB while `provider` is still live.
+    row.api_key = "sk-rotated"
+    integration_session.add(row)  # type: ignore[attr-defined]
+    await integration_session.commit()  # type: ignore[attr-defined]
+
+    with patch("routstr.proxy.reinitialize_upstreams", new=AsyncMock()):
+        await provider.on_upstream_error_redirect(402, "Insufficient balance")
+
+    await integration_session.refresh(row)  # type: ignore[attr-defined]
+    assert row.enabled is False
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_refresh_models_cache_finds_own_row_after_key_rotation(
+    integration_session: object,
+    patched_db_engine: None,
+) -> None:
+    """``refresh_models_cache`` must resolve its own row after a key rotation.
+
+    ``refresh_models_cache`` swallows every exception (it only logs), so the
+    observable proof it found its row is that it reaches ``list_models`` — which
+    is called with the row's ``id`` only *after* the row is resolved. RED
+    (current): the stale-key ``(base_url, api_key)`` lookup returns nothing, the
+    method raises ``404`` internally and returns before ``list_models`` is ever
+    called. GREEN: lookup by ``id`` finds the row and ``list_models`` runs for
+    that ``id``.
+    """
+    row = UpstreamProviderRow(
+        provider_type="ppqai",
+        base_url="https://api.ppq.ai",
+        api_key="sk-original",
+        enabled=True,
+        provider_fee=1.0,
+    )
+    integration_session.add(row)  # type: ignore[attr-defined]
+    await integration_session.commit()  # type: ignore[attr-defined]
+    await integration_session.refresh(row)  # type: ignore[attr-defined]
+    row_id = row.id
+
+    provider = PPQAIUpstreamProvider.from_db_row(row)  # captures sk-original
+    assert provider is not None
+
+    row.api_key = "sk-rotated"
+    integration_session.add(row)  # type: ignore[attr-defined]
+    await integration_session.commit()  # type: ignore[attr-defined]
+
+    list_models_mock = AsyncMock(return_value=[])
+    with (
+        patch.object(provider, "fetch_models", new=AsyncMock(return_value=[])),
+        patch("routstr.upstream.base.list_models", new=list_models_mock),
+    ):
+        await provider.refresh_models_cache()
+
+    list_models_mock.assert_awaited_once()
+    assert list_models_mock.await_args is not None
+    assert list_models_mock.await_args.kwargs["upstream_id"] == row_id

tests/integration/test_prune_dead_api_keys.py

@@ -0,0 +1,283 @@
+"""
+Tests for prune_dead_api_keys — the janitor that removes provably-dead 0/0/0
+API keys (funded keys fully refunded/expired without ever being used, plus bare
+orphans), while protecting keys that are still meaningful.
+"""
+
+import time
+import uuid
+from collections.abc import Awaitable, Callable
+from typing import Any, cast
+
+import pytest
+from sqlalchemy.sql.dml import Update
+from sqlmodel import col, update
+
+from routstr.core.db import (
+    ApiKey,
+    CashuTransaction,
+    LightningInvoice,
+    create_session,
+    prune_dead_api_keys,
+)
+
+OLD = 100  # min_age_seconds used by the tests
+NOW = int(time.time())
+LONG_AGO = NOW - 10_000  # well past the grace period
+
+
+async def _exists(key_hash: str) -> bool:
+    async with create_session() as session:
+        return (await session.get(ApiKey, key_hash)) is not None
+
+
+def _dead_key(created_at: int | None) -> ApiKey:
+    return ApiKey(
+        hashed_key=f"dead_{uuid.uuid4().hex}",
+        balance=0,
+        reserved_balance=0,
+        total_spent=0,
+        total_requests=0,
+        created_at=created_at,
+    )
+
+
+@pytest.mark.asyncio
+async def test_prunes_old_refunded_zero_key(patched_db_engine: None) -> None:
+    """A funded-then-refunded key (0/0/0, NULL parent, old) is pruned."""
+    key = _dead_key(LONG_AGO)
+    async with create_session() as session:
+        session.add(key)
+        await session.commit()
+
+    async with create_session() as session:
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 1
+    assert not await _exists(key.hashed_key)
+
+
+@pytest.mark.asyncio
+async def test_grace_period_protects_fresh_key(patched_db_engine: None) -> None:
+    """A dead-looking but recently created key is protected by the grace period."""
+    key = _dead_key(int(time.time()))
+    async with create_session() as session:
+        session.add(key)
+        await session.commit()
+
+    async with create_session() as session:
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 0
+    assert await _exists(key.hashed_key)
+
+
+@pytest.mark.asyncio
+async def test_used_key_never_pruned(patched_db_engine: None) -> None:
+    """Keys with any spend/requests or live balance are never pruned."""
+    spent = _dead_key(LONG_AGO)
+    spent.total_spent = 1
+    requested = _dead_key(LONG_AGO)
+    requested.total_requests = 1
+    funded = _dead_key(LONG_AGO)
+    funded.balance = 1000
+    reserved = _dead_key(LONG_AGO)
+    reserved.reserved_balance = 500
+
+    async with create_session() as session:
+        for k in (spent, requested, funded, reserved):
+            session.add(k)
+        await session.commit()
+
+    async with create_session() as session:
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 0
+    for k in (spent, requested, funded, reserved):
+        assert await _exists(k.hashed_key)
+
+
+@pytest.mark.asyncio
+async def test_parent_and_child_keys_are_not_pruned(
+    patched_db_engine: None,
+) -> None:
+    """Pruning must not orphan child keys or delete valid children."""
+    parent = _dead_key(LONG_AGO)
+    child = ApiKey(
+        hashed_key=f"child_{uuid.uuid4().hex}",
+        balance=0,
+        reserved_balance=0,
+        total_spent=0,
+        total_requests=0,
+        created_at=LONG_AGO,
+        parent_key_hash=parent.hashed_key,
+    )
+    async with create_session() as session:
+        session.add(parent)
+        session.add(child)
+        await session.commit()
+
+    async with create_session() as session:
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 0
+    assert await _exists(parent.hashed_key)
+    assert await _exists(child.hashed_key)
+
+
+@pytest.mark.asyncio
+async def test_pending_invoice_protects_key(patched_db_engine: None) -> None:
+    """A key referenced by a pending topup invoice is never pruned mid-topup."""
+    key = _dead_key(LONG_AGO)
+    invoice = LightningInvoice(
+        id=f"inv_{uuid.uuid4().hex}",
+        bolt11=f"lnbc_{uuid.uuid4().hex}",
+        amount_sats=10,
+        description="topup",
+        payment_hash=uuid.uuid4().hex,
+        status="pending",
+        api_key_hash=key.hashed_key,
+        purpose="topup",
+        expires_at=NOW + 10_000,
+    )
+    async with create_session() as session:
+        session.add(key)
+        session.add(invoice)
+        await session.commit()
+
+    async with create_session() as session:
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 0
+    assert await _exists(key.hashed_key)
+
+
+@pytest.mark.asyncio
+async def test_paid_invoice_does_not_protect_key(patched_db_engine: None) -> None:
+    """A settled (non-pending) invoice does not keep a dead key alive."""
+    key = _dead_key(LONG_AGO)
+    invoice = LightningInvoice(
+        id=f"inv_{uuid.uuid4().hex}",
+        bolt11=f"lnbc_{uuid.uuid4().hex}",
+        amount_sats=10,
+        description="topup",
+        payment_hash=uuid.uuid4().hex,
+        status="paid",
+        api_key_hash=key.hashed_key,
+        purpose="topup",
+        expires_at=NOW - 1,
+    )
+    async with create_session() as session:
+        session.add(key)
+        session.add(invoice)
+        await session.commit()
+
+    async with create_session() as session:
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 1
+    assert not await _exists(key.hashed_key)
+
+
+@pytest.mark.asyncio
+async def test_key_that_becomes_meaningful_during_prune_survives(
+    patched_db_engine: None, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """Revalidate before unlink/delete so a late top-up cannot be pruned."""
+    key = _dead_key(LONG_AGO)
+    txn = CashuTransaction(
+        id=uuid.uuid4().hex,
+        token="cashuABC",
+        amount=21,
+        unit="sat",
+        type="in",
+        source="apikey",
+        api_key_hashed_key=key.hashed_key,
+    )
+    async with create_session() as session:
+        session.add(key)
+        session.add(txn)
+        await session.commit()
+
+    async with create_session() as session:
+        original_exec = cast(Callable[..., Awaitable[Any]], session.exec)
+        topped_up = False
+
+        async def exec_with_late_topup(
+            statement: Any, *args: Any, **kwargs: Any
+        ) -> Any:
+            nonlocal topped_up
+            if (
+                not topped_up
+                and isinstance(statement, Update)
+                and getattr(statement.table, "name", None) == "cashu_transactions"
+            ):
+                topped_up = True
+                async with create_session() as topup_session:
+                    await topup_session.exec(  # type: ignore[call-overload]
+                        update(ApiKey)
+                        .where(col(ApiKey.hashed_key) == key.hashed_key)
+                        .values(balance=42)
+                    )
+                    await topup_session.commit()
+            return await original_exec(statement, *args, **kwargs)
+
+        monkeypatch.setattr(session, "exec", exec_with_late_topup)
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 0
+    assert await _exists(key.hashed_key)
+    async with create_session() as session:
+        surviving = await session.get(CashuTransaction, txn.id)
+        assert surviving is not None
+        assert surviving.api_key_hashed_key == key.hashed_key
+
+
+@pytest.mark.asyncio
+async def test_transaction_audit_trail_preserved(patched_db_engine: None) -> None:
+    """Pruning a refunded key keeps its cashu_transactions, unlinked from the key."""
+    key = _dead_key(LONG_AGO)
+    txn = CashuTransaction(
+        id=uuid.uuid4().hex,
+        token="cashuABC",
+        amount=21,
+        unit="sat",
+        type="in",
+        source="apikey",
+        api_key_hashed_key=key.hashed_key,
+    )
+    async with create_session() as session:
+        session.add(key)
+        session.add(txn)
+        await session.commit()
+
+    async with create_session() as session:
+        pruned = await prune_dead_api_keys(session, OLD)
+
+    assert pruned == 1
+    assert not await _exists(key.hashed_key)
+
+    async with create_session() as session:
+        surviving = await session.get(CashuTransaction, txn.id)
+        assert surviving is not None, "Financial audit row must survive key deletion"
+        assert surviving.api_key_hashed_key is None, "Link must be nulled, not dangling"
+        assert surviving.amount == 21
+
+
+@pytest.mark.asyncio
+async def test_periodic_prune_disabled_returns_immediately(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """Non-positive intervals disable the janitor."""
+    from unittest.mock import AsyncMock
+
+    from routstr import auth
+    from routstr.core.settings import settings
+
+    monkeypatch.setattr(settings, "dead_key_prune_interval_seconds", 0)
+    sleep_mock = AsyncMock()
+    monkeypatch.setattr(auth.asyncio, "sleep", sleep_mock)
+
+    await auth.periodic_dead_key_prune()
+
+    sleep_mock.assert_not_called()

tests/integration/test_swap_fee_retry.py

@@ -0,0 +1,194 @@
+"""
+Integration tests for reactive swap fee retries via the wallet topup endpoint.
+
+Foreign-mint tokens are swapped to the primary mint using the foreign mint's
+melt quote, whose fee_reserve is a non-binding estimate (NUT-05): the mint may
+demand more when re-quoting or at melt execution. These tests cover the
+endpoint behaviour in those cases:
+
+1. The mint demands one sat more at melt time than every quote reported
+   (the mint.cubabitcoin.org incident): the swap retries with a smaller
+   invoice and the topup succeeds, crediting the recomputed amount.
+2. The real melt quote reports a higher fee_reserve than the estimate: the
+   swap re-quotes from the observed fee and the topup succeeds.
+3. The mint escalates its fee demands on every attempt: the retry budget is
+   exhausted and the endpoint returns 400 with a clear error (never 500),
+   without ever executing a melt.
+"""
+
+from collections.abc import Callable
+from unittest.mock import AsyncMock, Mock, patch
+
+import pytest
+from httpx import AsyncClient, Response
+
+from routstr.core.settings import settings
+
+# Captured at collection time, before the integration_app fixture replaces it
+# with the testmint stub that bypasses swapping (see conftest.py).
+from routstr.wallet import recieve_token as _real_recieve_token
+
+PRIMARY_MINT = "http://primary:3338"
+
+
+def _make_swap_mocks(
+    token_amount: int,
+    fee_reserves: list[int],
+    input_fees: int = 0,
+    mint_url: str = "http://foreign-mint:3338",
+) -> tuple[Mock, Mock, Mock]:
+    """Return (token, token_wallet, primary_wallet) mocks that act like a mint.
+
+    Mint quotes pass the requested amount through their ``request`` field and
+    melt quotes echo that amount back, so the mocks stay consistent for
+    whatever amounts the implementation requests. ``fee_reserves`` supplies the
+    fee_reserve of each successive melt quote (the first serves the estimation
+    pass); requesting more quotes than provided fails the test.
+    """
+    mock_token = Mock()
+    mock_token.mint = mint_url
+    mock_token.unit = "sat"
+    mock_token.amount = token_amount
+    mock_token.keysets = ["keyset1"]
+    mock_token.proofs = [Mock(amount=token_amount)]
+
+    mock_token_wallet = Mock()
+    mock_token_wallet.load_mint = AsyncMock()
+    mock_token_wallet.load_proofs = AsyncMock()
+    mock_token_wallet.get_fees_for_proofs = Mock(return_value=input_fees)
+
+    mock_primary_wallet = Mock()
+    mock_primary_wallet.load_mint = AsyncMock()
+    mock_primary_wallet.load_proofs = AsyncMock()
+    mock_primary_wallet.available_balance = Mock(amount=0)
+    mock_primary_wallet.mint = AsyncMock(return_value=Mock())
+
+    fees = iter(fee_reserves)
+
+    def _next_fee() -> int:
+        try:
+            return next(fees)
+        except StopIteration:
+            raise AssertionError(
+                "more melt quotes requested than fee_reserves provided"
+            ) from None
+
+    mock_primary_wallet.request_mint = AsyncMock(
+        side_effect=lambda amount: Mock(quote=f"mint_quote_{amount}", request=amount)
+    )
+    mock_token_wallet.melt_quote = AsyncMock(
+        side_effect=lambda invoice: Mock(
+            quote=f"melt_quote_{invoice}", amount=invoice, fee_reserve=_next_fee()
+        )
+    )
+    mock_token_wallet.melt = AsyncMock(return_value=Mock())
+
+    return mock_token, mock_token_wallet, mock_primary_wallet
+
+
+def _wallet_router(primary_wallet: Mock, token_wallet: Mock) -> Callable[..., Mock]:
+    """Route get_wallet calls to the primary or foreign wallet mock by URL."""
+
+    def fake_get_wallet(mint_url: str, unit: str = "sat", load: bool = True) -> Mock:
+        return primary_wallet if mint_url == PRIMARY_MINT else token_wallet
+
+    return fake_get_wallet
+
+
+async def _post_topup(
+    client: AsyncClient,
+    mock_token: Mock,
+    token_wallet: Mock,
+    primary_wallet: Mock,
+) -> Response:
+    """POST /v1/wallet/topup with the swap layer mocked at the mint boundary.
+
+    The conftest's testmint stub for recieve_token is swapped back for the
+    real implementation so the request exercises the actual swap path.
+    """
+    with patch("routstr.wallet.recieve_token", _real_recieve_token):
+        with patch(
+            "routstr.wallet.deserialize_token_from_string", return_value=mock_token
+        ):
+            with patch(
+                "routstr.wallet.get_wallet",
+                side_effect=_wallet_router(primary_wallet, token_wallet),
+            ):
+                with patch.object(settings, "primary_mint", PRIMARY_MINT):
+                    with patch.object(settings, "primary_mint_unit", "sat"):
+                        with patch.object(settings, "cashu_mints", [PRIMARY_MINT]):
+                            return await client.post(
+                                "/v1/wallet/topup",
+                                params={"cashu_token": "cashuAtest_foreign_token"},
+                            )
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_topup_retries_when_melt_demands_more_than_quoted(
+    authenticated_client: AsyncClient,
+) -> None:
+    """A 179-sat token where every quote reports fee_reserve=1 but the mint
+    rejects the first melt demanding 180. The retry shrinks the invoice to 177
+    and the topup credits 177 sats (177_000 msats)."""
+    mock_token, token_wallet, primary_wallet = _make_swap_mocks(
+        179, fee_reserves=[1, 1, 1], mint_url="http://mint.cubabitcoin.org"
+    )
+    token_wallet.melt.side_effect = [
+        Exception(
+            "Mint Error: not enough inputs provided for melt. "
+            "Provided: 179, needed: 180 (Code: 11000)"
+        ),
+        Mock(),
+    ]
+
+    response = await _post_topup(
+        authenticated_client, mock_token, token_wallet, primary_wallet
+    )
+
+    assert response.status_code == 200
+    assert response.json()["msats"] == 177_000
+    assert token_wallet.melt.call_count == 2
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_topup_retries_when_quote_fee_exceeds_estimate(
+    authenticated_client: AsyncClient,
+) -> None:
+    """A 1000-sat token estimated at fee 20, but the real quote demands 23.
+    The retry recomputes 1000 - 23 = 977, which fits, and the topup credits
+    977 sats (977_000 msats) with a single melt."""
+    mock_token, token_wallet, primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[20, 23, 23]
+    )
+
+    response = await _post_topup(
+        authenticated_client, mock_token, token_wallet, primary_wallet
+    )
+
+    assert response.status_code == 200
+    assert response.json()["msats"] == 977_000
+    assert token_wallet.melt.call_count == 1
+
+
+@pytest.mark.integration
+@pytest.mark.asyncio
+async def test_topup_returns_400_when_retries_exhausted(
+    authenticated_client: AsyncClient,
+) -> None:
+    """A mint that escalates fee_reserve on every re-quote (1 → 10 → 25 → 50)
+    exhausts the retry budget: clean 400 with an actionable message, melt never
+    executed."""
+    mock_token, token_wallet, primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[1, 10, 25, 50]
+    )
+
+    response = await _post_topup(
+        authenticated_client, mock_token, token_wallet, primary_wallet
+    )
+
+    assert response.status_code == 400
+    assert "too small to cover swap fees" in response.json()["detail"]
+    assert token_wallet.melt_quote.call_count == 4  # estimation + 3 attempts
+    token_wallet.melt.assert_not_called()

tests/unit/test_balance.py

@@ -2,6 +2,7 @@ import json
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
+from fastapi import HTTPException
 from fastapi.responses import JSONResponse
 
 from routstr.balance import refund_wallet_endpoint
@@ -88,8 +89,6 @@ async def test_refund_x_cashu_sat_unit() -> None:
 
 @pytest.mark.asyncio
 async def test_refund_x_cashu_not_found_raises_404() -> None:
-    from fastapi import HTTPException
-
     session = MagicMock()
     session.exec = AsyncMock(return_value=_exec_result(None))
 
@@ -104,9 +103,32 @@ async def test_refund_x_cashu_not_found_raises_404() -> None:
 
 
 @pytest.mark.asyncio
-async def test_refund_x_cashu_swept_raises_410() -> None:
-    from fastapi import HTTPException
+async def test_refund_x_cashu_pending_out_tx_raises_425() -> None:
+    in_tx = _make_cashu_tx(
+        token="cashuApending_token",
+        amount=0,
+        unit="msat",
+        type="in",
+        request_id="req-pending",
+    )
 
+    session = MagicMock()
+    session.exec = AsyncMock(side_effect=[_exec_result(in_tx), _exec_result(None)])
+
+    with pytest.raises(HTTPException) as exc_info:
+        await refund_wallet_endpoint(
+            authorization="Bearer sk-somekey",
+            x_cashu="cashuApending_token",
+            session=session,
+        )
+
+    assert exc_info.value.status_code == 425
+    assert exc_info.value.detail == "Refund not ready. Retry later."
+    assert exc_info.value.headers == {"Retry-After": "5"}
+
+
+@pytest.mark.asyncio
+async def test_refund_x_cashu_swept_raises_410() -> None:
     in_tx = _make_cashu_tx(token="cashuAswept_token", amount=0, unit="msat", type="in", request_id="req-swept")
     out_tx = _make_cashu_tx(token="cashuAswept", amount=100, unit="msat", type="out", request_id="req-swept", swept=True)
 

tests/unit/test_cache_pricing.py

@@ -81,6 +81,21 @@ def test_backfill_strips_vendor_prefix_for_litellm_lookup() -> None:
     assert result.input_cache_read == expected
 
 
+def test_backfill_case_insensitive_lookup() -> None:
+    """A generic upstream may report a mixed-case id
+    (deepseek-ai/DeepSeek-V4-Flash); litellm keys are lowercase. The
+    case-insensitive fallback still resolves the cache rate."""
+    pricing = Pricing(prompt=1.4e-07, completion=2.8e-07)
+
+    result = backfill_cache_pricing("deepseek-ai/DeepSeek-V4-Flash", pricing)
+
+    expected = litellm.model_cost["deepseek-v4-flash"][
+        "cache_read_input_token_cost"
+    ]
+    assert result.input_cache_read == expected
+    assert result.input_cache_read < pricing.prompt  # sanity: it's a discount
+
+
 def test_backfill_fills_cache_write_rate() -> None:
     """Anthropic cache writes cost more than input (1.25x); billing them at
     the input rate undercharges. litellm carries the write rate."""
@@ -136,6 +151,93 @@ def test_provider_fee_applies_to_backfilled_cache_rates() -> None:
     assert adjusted.pricing.prompt == pytest.approx(2.8e-07 * 2.0)
 
 
+def test_row_to_model_backfills_cache_rate() -> None:
+    """The DB-override path (admin-configured providers, e.g. a generic
+    upstream) stores pricing without cache rates. ``_row_to_model`` must
+    backfill them from litellm just like ``_apply_provider_fee_to_model``,
+    otherwise cache reads bill at the full input rate."""
+    import json
+
+    from routstr.core.db import ModelRow
+    from routstr.payment.models import _row_to_model
+
+    row = ModelRow(
+        id="deepseek-v4-flash",
+        name="deepseek-v4-flash",
+        created=0,
+        description="",
+        context_length=1000000,
+        architecture=json.dumps(
+            {
+                "modality": "text",
+                "input_modalities": ["text"],
+                "output_modalities": ["text"],
+                "tokenizer": "unknown",
+                "instruct_type": None,
+            }
+        ),
+        # Stored pricing omits input_cache_read (generic provider never sets it).
+        pricing=json.dumps({"prompt": 1.4e-07, "completion": 2.8e-07}),
+        enabled=True,
+        upstream_provider_id=1,
+    )
+
+    with patch(
+        "routstr.payment.models.sats_usd_price", return_value=5.0e-5
+    ):
+        model = _row_to_model(row, apply_provider_fee=True, provider_fee=1.0)
+
+    litellm_read = litellm.model_cost["deepseek-v4-flash"][
+        "cache_read_input_token_cost"
+    ]
+    assert model.pricing.input_cache_read == pytest.approx(litellm_read)
+    assert model.pricing.input_cache_read < model.pricing.prompt  # a discount
+    assert model.sats_pricing is not None
+    assert model.sats_pricing.input_cache_read > 0
+
+
+def test_row_to_model_backfills_via_forwarded_model_id() -> None:
+    """An alias row (id != forwarded_model_id) must backfill cache rates from
+    the *forwarded* model name — the real upstream model litellm prices —
+    not the alias id, which litellm doesn't know."""
+    import json
+
+    from routstr.core.db import ModelRow
+    from routstr.payment.models import _row_to_model
+
+    row = ModelRow(
+        id="local-alias",  # litellm has no such key
+        name="local-alias",
+        created=0,
+        description="",
+        context_length=1000000,
+        architecture=json.dumps(
+            {
+                "modality": "text",
+                "input_modalities": ["text"],
+                "output_modalities": ["text"],
+                "tokenizer": "unknown",
+                "instruct_type": None,
+            }
+        ),
+        pricing=json.dumps({"prompt": 1.4e-07, "completion": 2.8e-07}),
+        enabled=True,
+        upstream_provider_id=1,
+        forwarded_model_id="deepseek-v4-flash",
+    )
+
+    with patch(
+        "routstr.payment.models.sats_usd_price", return_value=5.0e-5
+    ):
+        model = _row_to_model(row, apply_provider_fee=True, provider_fee=1.0)
+
+    litellm_read = litellm.model_cost["deepseek-v4-flash"][
+        "cache_read_input_token_cost"
+    ]
+    assert model.pricing.input_cache_read == pytest.approx(litellm_read)
+    assert model.pricing.input_cache_read < model.pricing.prompt
+
+
 # ============================================================================
 # calculate_cost — cached tokens billed at cache rates
 # ============================================================================

tests/unit/test_cost_calculation_caching.py

@@ -404,6 +404,67 @@ async def test_deepseek_malformed_hit_tokens_coerce_to_zero() -> None:
     assert result.cache_read_input_tokens == 0
 
 
+# ============================================================================
+# Truly-empty response with a non-zero USD cost → full refund
+#
+# When an upstream reports a USD cost but the response carries NO tokens at all
+# (input, output, cache-read and cache-creation all zero), billing the
+# USD-derived cost charges the user for nothing. Refund in full. The gate is
+# tightened relative to PR #489: a cache-read/-creation-only turn legitimately
+# reports zero prompt/completion tokens with a real cost and must still bill.
+# ============================================================================
+@pytest.mark.asyncio
+async def test_truly_empty_usd_cost_response_is_refunded(
+    mock_fixed_pricing: None,
+) -> None:
+    """0 input + 0 output + 0 cache tokens with a non-zero USD cost → refund."""
+    response = {
+        "model": "gpt-4",
+        "usage": {
+            "prompt_tokens": 0,
+            "completion_tokens": 0,
+            "total_cost": 0.01,  # non-zero USD cost despite no tokens
+        },
+    }
+    result = await calculate_cost(response, max_cost=100000)
+
+    assert isinstance(result, CostData)
+    assert result.total_msats == 0  # full refund
+    assert result.input_msats == 0
+    assert result.output_msats == 0
+    assert result.total_usd == 0.0
+    assert result.input_tokens == 0
+    assert result.output_tokens == 0
+    assert result.cache_read_input_tokens == 0
+    assert result.cache_creation_input_tokens == 0
+
+
+@pytest.mark.asyncio
+async def test_cache_read_only_usd_cost_response_is_billed(
+    mock_fixed_pricing: None,
+) -> None:
+    """Cache-read-only turn (0 prompt/completion, non-zero cost) still bills."""
+    response = {
+        "model": "claude-3-5-sonnet",
+        "usage": {
+            "input_tokens": 0,
+            "output_tokens": 0,
+            "cache_read_input_tokens": 1000,  # real cached usage
+            "cache_creation_input_tokens": 0,
+            "total_cost": 0.01,  # non-zero USD cost
+        },
+    }
+    result = await calculate_cost(response, max_cost=100000)
+
+    assert isinstance(result, CostData)
+    # NOT refunded — the USD cost is billed in full. Pinning the exact value
+    # guards against any future regression that would over-refund a cache-only
+    # turn (the bug in PR #489, which refunded whenever prompt+completion == 0).
+    assert result.total_msats == 200000
+    assert result.total_usd == 0.01
+    assert result.cache_read_input_tokens == 1000
+
+
 # ============================================================================
 # Test 13: Missing Usage Block
 # ============================================================================

tests/unit/test_provider_slug_migration.py

@@ -0,0 +1,55 @@
+from __future__ import annotations
+
+import importlib.util
+from pathlib import Path
+
+import sqlalchemy as sa
+
+_MIGRATION_PATH = (
+    Path(__file__).resolve().parents[2]
+    / "migrations"
+    / "versions"
+    / "c6d7e8f9a0b1_add_slug_to_upstream_providers.py"
+)
+_spec = importlib.util.spec_from_file_location("provider_slug_migration", _MIGRATION_PATH)
+assert _spec is not None and _spec.loader is not None
+migration = importlib.util.module_from_spec(_spec)
+_spec.loader.exec_module(migration)
+
+
+def test_slug_migration_backfill_uses_api_safe_deterministic_slugs() -> None:
+    engine = sa.create_engine("sqlite:///:memory:")
+    with engine.begin() as conn:
+        conn.execute(
+            sa.text(
+                "CREATE TABLE upstream_providers ("
+                "id INTEGER PRIMARY KEY, "
+                "provider_type VARCHAR NOT NULL, "
+                "slug VARCHAR NULL"
+                ")"
+            )
+        )
+        conn.execute(
+            sa.text(
+                "INSERT INTO upstream_providers (id, provider_type, slug) VALUES "
+                "(1, 'OpenAI Compatible', NULL), "
+                "(2, 'OpenAI Compatible', ''), "
+                "(3, '123', NULL), "
+                "(4, 'x', NULL), "
+                "(5, 'anthropic', 'anthropic')"
+            )
+        )
+
+        migration._backfill_provider_slugs(conn)
+
+        rows = conn.execute(
+            sa.text("SELECT id, slug FROM upstream_providers ORDER BY id")
+        ).all()
+
+    assert rows == [
+        (1, "openai-compatible"),
+        (2, "openai-compatible-2"),
+        (3, "provider-123"),
+        (4, "x-provider"),
+        (5, "anthropic"),
+    ]

tests/unit/test_provider_slugs.py

@@ -0,0 +1,144 @@
+from __future__ import annotations
+
+import pytest
+from sqlalchemy.ext.asyncio import create_async_engine
+from sqlmodel import SQLModel, select
+from sqlmodel.ext.asyncio.session import AsyncSession
+
+from routstr.core.admin import _get_upstream_provider_by_ref
+from routstr.core.db import UpstreamProviderRow
+from routstr.core.provider_slugs import (
+    allocate_unique_provider_slug,
+    provider_slug_base,
+)
+from routstr.upstream.helpers import _seed_providers_from_settings
+
+
+@pytest.mark.asyncio
+async def test_allocate_unique_provider_slug_is_deterministic_with_suffixes() -> None:
+    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
+    async with engine.begin() as conn:
+        await conn.run_sync(SQLModel.metadata.create_all)
+
+    async with AsyncSession(engine) as session:
+        session.add(
+            UpstreamProviderRow(
+                slug="openai",
+                provider_type="openai",
+                base_url="https://api.openai.com/v1",
+                api_key="key-1",
+            )
+        )
+        await session.commit()
+
+        assert await allocate_unique_provider_slug(session, "openai") == "openai-2"
+        assert (
+            await allocate_unique_provider_slug(session, "openai", {"openai-2"})
+            == "openai-3"
+        )
+
+    await engine.dispose()
+
+
+def test_provider_slug_base_sanitizes_provider_type() -> None:
+    assert provider_slug_base("OpenAI Compatible") == "openai-compatible"
+    assert provider_slug_base("!!!") == "provider"
+    assert provider_slug_base("AI") == "ai-provider"
+    assert provider_slug_base("123") == "provider-123"
+
+
+@pytest.mark.asyncio
+async def test_provider_ref_lookup_accepts_existing_numeric_ids_and_slugs() -> None:
+    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
+    async with engine.begin() as conn:
+        await conn.run_sync(SQLModel.metadata.create_all)
+
+    async with AsyncSession(engine) as session:
+        provider = UpstreamProviderRow(
+            slug="openai",
+            provider_type="openai",
+            base_url="https://api.openai.com/v1",
+            api_key="key-1",
+        )
+        session.add(provider)
+        await session.commit()
+        await session.refresh(provider)
+
+        by_id = await _get_upstream_provider_by_ref(session, str(provider.id))
+        by_slug = await _get_upstream_provider_by_ref(session, "openai")
+
+    assert by_id.id == provider.id
+    assert by_slug.id == provider.id
+
+    await engine.dispose()
+
+
+@pytest.mark.asyncio
+async def test_seed_providers_from_settings_sets_deterministic_slug(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
+    async with engine.begin() as conn:
+        await conn.run_sync(SQLModel.metadata.create_all)
+
+    monkeypatch.setenv("OPENAI_API_KEY", "seeded-openai-key")
+
+    class SettingsStub:
+        chat_completions_api_version: str | None = None
+        upstream_base_url: str | None = None
+        upstream_api_key: str = ""
+
+    async with AsyncSession(engine) as session:
+        await _seed_providers_from_settings(session, SettingsStub())  # type: ignore[arg-type]
+        await session.commit()
+
+        result = await session.exec(select(UpstreamProviderRow))
+        providers: list[UpstreamProviderRow] = list(result.all())
+
+    assert [(p.provider_type, p.slug) for p in providers] == [("openai", "openai")]
+
+    await engine.dispose()
+
+
+@pytest.mark.asyncio
+async def test_seed_providers_from_settings_keeps_slug_stable_on_reseed(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    engine = create_async_engine("sqlite+aiosqlite:///:memory:")
+    async with engine.begin() as conn:
+        await conn.run_sync(SQLModel.metadata.create_all)
+
+    monkeypatch.setenv("OPENAI_API_KEY", "seeded-openai-key")
+
+    class SettingsStub:
+        chat_completions_api_version: str | None = None
+        upstream_base_url: str | None = None
+        upstream_api_key: str = ""
+
+    async with AsyncSession(engine) as session:
+        session.add(
+            UpstreamProviderRow(
+                slug="openai",
+                provider_type="openai",
+                base_url="https://example.invalid/v1",
+                api_key="other-key",
+            )
+        )
+        await session.commit()
+
+        await _seed_providers_from_settings(session, SettingsStub())  # type: ignore[arg-type]
+        await session.commit()
+        await _seed_providers_from_settings(session, SettingsStub())  # type: ignore[arg-type]
+        await session.commit()
+
+        result = await session.exec(
+            select(UpstreamProviderRow).order_by(UpstreamProviderRow.slug)
+        )
+        providers: list[UpstreamProviderRow] = list(result.all())
+
+    assert [(p.provider_type, p.slug) for p in providers] == [
+        ("openai", "openai"),
+        ("openai", "openai-2"),
+    ]
+
+    await engine.dispose()

tests/unit/test_upstream_generic.py

@@ -0,0 +1,524 @@
+"""Unit tests for ``GenericUpstreamProvider.fetch_models`` price/metadata resolution.
+
+A generic upstream is any OpenAI-compatible API. Most (DeepSeek, OpenAI,
+Groq, ...) answer ``/models`` with bare ``{id, object, owned_by}`` entries that
+carry *no* pricing. The provider must not fabricate a price for those: it
+resolves through native ``model_spec`` (Venice's bespoke schema) → litellm's
+bundled cost map → the OpenRouter feed, and only when every source misses does
+it import the model **disabled** with a warning rather than invent a number.
+
+These tests drive that behaviour through the public ``fetch_models`` API. The
+``/models`` HTTP call is faked at ``httpx.AsyncClient``; the OpenRouter feed is
+patched at its source (``routstr.payment.models.async_fetch_openrouter_models``)
+so the resolver's lazy import picks up the stub. litellm's real bundled cost map
+is used unmocked — the DeepSeek rates it ships are the assertion's ground truth.
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from routstr.upstream.generic import GenericUpstreamProvider
+
+
+class _FakeResponse:
+    def __init__(self, payload: dict[str, Any]) -> None:
+        self._payload = payload
+
+    def raise_for_status(self) -> None:
+        return None
+
+    def json(self) -> dict[str, Any]:
+        return self._payload
+
+
+class _FakeAsyncClient:
+    """Stand-in for ``httpx.AsyncClient`` returning a canned ``/models`` body."""
+
+    def __init__(self, payload: dict[str, Any]) -> None:
+        self._payload = payload
+
+    async def __aenter__(self) -> "_FakeAsyncClient":
+        return self
+
+    async def __aexit__(self, *exc: object) -> bool:
+        return False
+
+    async def get(self, url: str, headers: dict[str, str] | None = None) -> _FakeResponse:
+        return _FakeResponse(self._payload)
+
+
+def _patch_models_endpoint(payload: dict[str, Any]) -> Any:
+    """Patch the provider's ``httpx.AsyncClient`` to serve ``payload``."""
+    return patch(
+        "routstr.upstream.generic.httpx.AsyncClient",
+        lambda *args, **kwargs: _FakeAsyncClient(payload),
+    )
+
+
+def _model_by_id(models: list[Any], model_id: str) -> Any:
+    return next(m for m in models if m.id == model_id)
+
+
+# ---------------------------------------------------------------------------
+# native model_spec (Venice) — must keep resolving, and capture its metadata
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_native_model_spec_resolves_and_captures_metadata() -> None:
+    """A Venice-shaped ``model_spec`` is authoritative: prices/context come
+    straight from it and vision capability becomes an image input modality."""
+    payload = {
+        "data": [
+            {
+                "id": "venice-llama",
+                "owned_by": "venice",
+                "model_spec": {
+                    "name": "Venice Llama",
+                    "availableContextTokens": 65536,
+                    "pricing": {
+                        "input": {"usd": 0.5},
+                        "output": {"usd": 1.5},
+                    },
+                    "capabilities": {"supportsVision": True},
+                },
+            }
+        ]
+    }
+
+    with _patch_models_endpoint(payload):
+        or_feed = AsyncMock(return_value=[])
+        with patch(
+            "routstr.payment.models.async_fetch_openrouter_models", or_feed
+        ):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "venice-llama")
+    assert model.enabled is True
+    assert model.pricing.prompt == pytest.approx(0.5 / 1_000_000)
+    assert model.pricing.completion == pytest.approx(1.5 / 1_000_000)
+    assert model.context_length == 65536
+    assert "image" in model.architecture.input_modalities
+    # Vision capability must be reflected in the combined modality string, not
+    # flattened to "text->text".
+    assert model.architecture.modality == "text+image->text"
+    # A native price never needs the OpenRouter feed.
+    or_feed.assert_not_awaited()
+
+
+# ---------------------------------------------------------------------------
+# native model_spec validation — a bogus native price is not authoritative
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_native_both_zero_price_falls_through_to_litellm() -> None:
+    """A native ``model_spec`` that prices both tokens at 0 is not a real price
+    (the same free-tier trap the litellm/OpenRouter rungs already reject). It
+    must not be treated as authoritative and served free; the resolver falls
+    through, so a litellm-known model lands on litellm's real rate instead."""
+    payload = {
+        "data": [
+            {
+                "id": "deepseek-chat",
+                "owned_by": "deepseek",
+                "model_spec": {
+                    "pricing": {"input": {"usd": 0}, "output": {"usd": 0}},
+                },
+            }
+        ]
+    }
+
+    with _patch_models_endpoint(payload):
+        or_feed = AsyncMock(return_value=[])
+        with patch("routstr.payment.models.async_fetch_openrouter_models", or_feed):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "deepseek-chat")
+    assert model.enabled is True
+    assert model.pricing.prompt == pytest.approx(2.8e-07)
+    assert model.pricing.completion == pytest.approx(4.2e-07)
+
+
+@pytest.mark.asyncio
+async def test_native_negative_price_falls_through_to_litellm() -> None:
+    """A negative native price would credit the caller's balance on every
+    request (a fund drain, not a discount). Reject it like any other unusable
+    price and fall through to the chain."""
+    payload = {
+        "data": [
+            {
+                "id": "deepseek-chat",
+                "owned_by": "deepseek",
+                "model_spec": {
+                    "pricing": {"input": {"usd": -0.5}, "output": {"usd": -1.5}},
+                },
+            }
+        ]
+    }
+
+    with _patch_models_endpoint(payload):
+        or_feed = AsyncMock(return_value=[])
+        with patch("routstr.payment.models.async_fetch_openrouter_models", or_feed):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "deepseek-chat")
+    assert model.enabled is True
+    assert model.pricing.prompt == pytest.approx(2.8e-07)
+    assert model.pricing.completion == pytest.approx(4.2e-07)
+
+
+@pytest.mark.asyncio
+async def test_native_non_numeric_price_does_not_break_catalog() -> None:
+    """A non-numeric native price (``"free"``) must not raise while parsing —
+    an unguarded ``"free" / 1_000_000`` throws and the outer catch drops the
+    *entire* provider catalog. It has to fail closed for that one model while
+    every other model in the same response still resolves."""
+    payload = {
+        "data": [
+            {
+                "id": "broken-price",
+                "owned_by": "mystery",
+                "model_spec": {
+                    "pricing": {"input": {"usd": "free"}, "output": {"usd": "free"}},
+                },
+            },
+            {"id": "deepseek-chat", "object": "model", "owned_by": "deepseek"},
+        ]
+    }
+
+    with _patch_models_endpoint(payload):
+        or_feed = AsyncMock(return_value=[])
+        with patch("routstr.payment.models.async_fetch_openrouter_models", or_feed):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    # One malformed entry must not empty the catalog.
+    assert {m.id for m in models} == {"broken-price", "deepseek-chat"}
+    broken = _model_by_id(models, "broken-price")
+    assert broken.enabled is False
+    healthy = _model_by_id(models, "deepseek-chat")
+    assert healthy.enabled is True
+    assert healthy.pricing.prompt == pytest.approx(2.8e-07)
+
+
+# ---------------------------------------------------------------------------
+# litellm rescue — the money-critical case (DeepSeek bare /models)
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_bare_deepseek_resolves_via_litellm() -> None:
+    """DeepSeek's ``/models`` carries no pricing. The old code fabricated
+    ``$0.001`` + ctx 4096; the resolver must instead pull DeepSeek's real
+    rates from litellm's bundled cost map (``$0.28``/``$0.42`` per 1M, ctx
+    131072) and keep the model enabled."""
+    payload = {
+        "data": [
+            {"id": "deepseek-chat", "object": "model", "owned_by": "deepseek"},
+        ]
+    }
+
+    with _patch_models_endpoint(payload):
+        or_feed = AsyncMock(return_value=[])
+        with patch(
+            "routstr.payment.models.async_fetch_openrouter_models", or_feed
+        ):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "deepseek-chat")
+    assert model.enabled is True
+    assert model.pricing.prompt == pytest.approx(2.8e-07)
+    assert model.pricing.completion == pytest.approx(4.2e-07)
+    assert model.context_length == 131072
+    # Richer metadata than the two base prices is captured too.
+    assert model.pricing.input_cache_read == pytest.approx(2.8e-08)
+    assert model.top_provider is not None
+    assert model.top_provider.max_completion_tokens == 8192
+    # litellm answered, so the OpenRouter feed is never consulted.
+    or_feed.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_litellm_zero_price_entry_fails_closed(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """A litellm entry that lists a model but prices it at 0/0 (free-tier
+    moderation/rerank models do this) is not a real price — treating it as one
+    would silently serve the model for free. The resolver must reject a both-zero
+    litellm hit and fall through, so the model imports disabled, not at $0."""
+    payload = {
+        "data": [
+            {"id": "omni-moderation-latest", "object": "model", "owned_by": "openai"},
+        ]
+    }
+
+    gen_logger = logging.getLogger("routstr.upstream.generic")
+    gen_logger.addHandler(caplog.handler)
+    try:
+        with _patch_models_endpoint(payload):
+            or_feed = AsyncMock(return_value=[])
+            with patch(
+                "routstr.payment.models.async_fetch_openrouter_models", or_feed
+            ):
+                models = await GenericUpstreamProvider(
+                    base_url="http://x"
+                ).fetch_models()
+    finally:
+        gen_logger.removeHandler(caplog.handler)
+
+    model = _model_by_id(models, "omni-moderation-latest")
+    assert model.enabled is False
+    assert model.pricing.prompt == 0.0
+    assert model.pricing.completion == 0.0
+    assert any(
+        "omni-moderation-latest" in rec.getMessage()
+        for rec in caplog.records
+        if rec.levelno >= logging.WARNING
+    )
+
+
+@pytest.mark.asyncio
+async def test_litellm_output_cap_not_used_as_context() -> None:
+    """litellm's ``max_tokens`` is the completion cap, not the context window
+    (it tracks ``max_output_tokens`` for ~94% of models). When a model reports
+    no ``max_input_tokens``, the resolver must not smuggle the output cap in as
+    the context window; it falls back to the id-based estimate instead, while
+    ``max_tokens`` still feeds the completion limit."""
+    payload = {
+        "data": [
+            {
+                "id": "gemini/gemini-gemma-2-9b-it",
+                "object": "model",
+                "owned_by": "google",
+            },
+        ]
+    }
+
+    with _patch_models_endpoint(payload):
+        or_feed = AsyncMock(return_value=[])
+        with patch(
+            "routstr.payment.models.async_fetch_openrouter_models", or_feed
+        ):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "gemini/gemini-gemma-2-9b-it")
+    assert model.enabled is True
+    # litellm gives this model max_input_tokens=None, max_tokens=8192 (an output
+    # cap). Context must come from the estimate (4096), never the 8192 cap.
+    assert model.context_length == 4096
+    # The 8192 output cap still lands where it belongs: the completion limit.
+    assert model.top_provider is not None
+    assert model.top_provider.max_completion_tokens == 8192
+
+
+# ---------------------------------------------------------------------------
+# OpenRouter fallback — litellm misses, OR carries a full payload
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_unknown_to_litellm_resolves_via_openrouter() -> None:
+    """A model litellm has never heard of still resolves if the OpenRouter
+    feed lists it, pulling price + context from that entry."""
+    payload = {
+        "data": [
+            {"id": "exotic/model-9000", "object": "model", "owned_by": "exotic"},
+        ]
+    }
+    or_entry = {
+        "id": "exotic/model-9000",
+        "name": "Exotic 9000",
+        "context_length": 65536,
+        "architecture": {
+            "modality": "text+image->text",
+            "input_modalities": ["text", "image"],
+            "output_modalities": ["text"],
+            "tokenizer": "Other",
+            "instruct_type": None,
+        },
+        "pricing": {"prompt": "0.000005", "completion": "0.000010"},
+        "top_provider": {
+            "context_length": 65536,
+            "max_completion_tokens": 4096,
+            "is_moderated": False,
+        },
+    }
+
+    with _patch_models_endpoint(payload):
+        or_feed = AsyncMock(return_value=[or_entry])
+        with patch(
+            "routstr.payment.models.async_fetch_openrouter_models", or_feed
+        ):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "exotic/model-9000")
+    assert model.enabled is True
+    assert model.pricing.prompt == pytest.approx(5e-06)
+    assert model.pricing.completion == pytest.approx(1e-05)
+    assert model.context_length == 65536
+    # The feed's own modality string is carried through verbatim, not recomputed.
+    assert model.architecture.modality == "text+image->text"
+    or_feed.assert_awaited()
+
+
+@pytest.mark.asyncio
+async def test_openrouter_bare_tail_collision_picks_highest_price() -> None:
+    """When a bare model id matches several OpenRouter entries by tail
+    (``model`` ↔ ``a/model``, ``b/model``), the match must be deterministic and
+    money-safe: pick the highest-priced candidate regardless of feed order, so
+    ordering can never leave the node charging below true cost. (The live feed
+    has zero such collisions today; this guards the latent case.)"""
+    payload = {
+        "data": [
+            {"id": "zzz-phantom-model", "object": "model", "owned_by": "mystery"},
+        ]
+    }
+    # Same bare tail, different resellers; the pricier one is listed *second*
+    # so a first-wins match would pick the cheaper (undercharging) entry.
+    or_feed = AsyncMock(
+        return_value=[
+            {
+                "id": "cheapco/zzz-phantom-model",
+                "context_length": 8192,
+                "pricing": {"prompt": "0.000001", "completion": "0.000002"},
+            },
+            {
+                "id": "premiumco/zzz-phantom-model",
+                "context_length": 8192,
+                "pricing": {"prompt": "0.000009", "completion": "0.000010"},
+            },
+        ]
+    )
+
+    with _patch_models_endpoint(payload):
+        with patch("routstr.payment.models.async_fetch_openrouter_models", or_feed):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "zzz-phantom-model")
+    assert model.pricing.prompt == pytest.approx(9e-06)
+    assert model.pricing.completion == pytest.approx(1e-05)
+
+
+@pytest.mark.asyncio
+async def test_openrouter_bare_tail_tie_breaks_on_combined_cost() -> None:
+    """The bare-tail tie-break must weigh *both* rates, not prompt alone.
+    Given two colliding entries where one is cheaper on prompt but far dearer
+    on completion, ranking by prompt would pick the entry that undercharges
+    output-heavy traffic. Pick the highest *combined* per-token cost so the
+    money-safe choice holds whichever way the traffic leans."""
+    payload = {
+        "data": [
+            {"id": "yyy-phantom-model", "object": "model", "owned_by": "mystery"},
+        ]
+    }
+    # dear-overall is listed first with the *lower* prompt, so a prompt-only max
+    # would wrongly pick the second (cheaper-overall) entry.
+    or_feed = AsyncMock(
+        return_value=[
+            {
+                "id": "dearco/yyy-phantom-model",
+                "context_length": 8192,
+                "pricing": {"prompt": "0.000001", "completion": "0.000100"},
+            },
+            {
+                "id": "cheapco/yyy-phantom-model",
+                "context_length": 8192,
+                "pricing": {"prompt": "0.000009", "completion": "0.000002"},
+            },
+        ]
+    )
+
+    with _patch_models_endpoint(payload):
+        with patch("routstr.payment.models.async_fetch_openrouter_models", or_feed):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    model = _model_by_id(models, "yyy-phantom-model")
+    assert model.pricing.prompt == pytest.approx(1e-06)
+    assert model.pricing.completion == pytest.approx(1e-04)
+
+
+@pytest.mark.asyncio
+async def test_openrouter_feed_fetched_once_per_discovery() -> None:
+    """Two models both missing litellm must share a single OpenRouter fetch —
+    the feed is not re-downloaded per model."""
+    payload = {
+        "data": [
+            {"id": "exotic/model-a", "object": "model", "owned_by": "exotic"},
+            {"id": "exotic/model-b", "object": "model", "owned_by": "exotic"},
+        ]
+    }
+    or_feed = AsyncMock(
+        return_value=[
+            {
+                "id": "exotic/model-a",
+                "context_length": 8192,
+                "pricing": {"prompt": "0.000001", "completion": "0.000002"},
+            },
+            {
+                "id": "exotic/model-b",
+                "context_length": 8192,
+                "pricing": {"prompt": "0.000003", "completion": "0.000004"},
+            },
+        ]
+    )
+
+    with _patch_models_endpoint(payload):
+        with patch("routstr.payment.models.async_fetch_openrouter_models", or_feed):
+            models = await GenericUpstreamProvider(base_url="http://x").fetch_models()
+
+    assert {m.id for m in models} == {"exotic/model-a", "exotic/model-b"}
+    assert or_feed.await_count == 1
+
+
+# ---------------------------------------------------------------------------
+# fail closed — no source resolves → disabled + warned, never fabricated
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_unresolvable_model_fails_closed(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """When native, litellm and OpenRouter all miss, the model is imported
+    disabled with a warning naming it — and no price is invented (the old
+    ``$0.001`` placeholder is gone)."""
+    payload = {
+        "data": [
+            {
+                "id": "nobody-has-priced-this-xyz",
+                "object": "model",
+                "owned_by": "mystery",
+            },
+        ]
+    }
+
+    # routstr loggers set propagate=False, so caplog's root handler misses
+    # them; attach its handler to the provider logger directly.
+    gen_logger = logging.getLogger("routstr.upstream.generic")
+    gen_logger.addHandler(caplog.handler)
+    try:
+        with _patch_models_endpoint(payload):
+            or_feed = AsyncMock(return_value=[])
+            with patch(
+                "routstr.payment.models.async_fetch_openrouter_models", or_feed
+            ):
+                models = await GenericUpstreamProvider(
+                    base_url="http://x"
+                ).fetch_models()
+    finally:
+        gen_logger.removeHandler(caplog.handler)
+
+    model = _model_by_id(models, "nobody-has-priced-this-xyz")
+    assert model.enabled is False
+    assert model.pricing.prompt == 0.0
+    assert model.pricing.completion == 0.0
+    assert any(
+        "nobody-has-priced-this-xyz" in rec.getMessage()
+        for rec in caplog.records
+        if rec.levelno >= logging.WARNING
+    )

tests/unit/test_upstream_rate_limit.py

@@ -0,0 +1,396 @@
+"""Tests for upstream rate-limit detection, classification, and org-ID redaction.
+
+Covers issue #555: upstream OpenAI-compatible providers return rate-limit
+errors that embed a sensitive organization ID. The proxy must classify these
+distinctly (``UPSTREAM_RATE_LIMIT``), preserve useful debugging fields, and
+never emit a raw ``org-*`` identifier in logs, errors, or returned bodies.
+"""
+
+from __future__ import annotations
+
+import json
+from typing import Any
+from unittest.mock import AsyncMock, MagicMock, Mock, patch
+
+import httpx
+import pytest
+
+from routstr.core.redaction import redact_org_ids
+from routstr.upstream.base import BaseUpstreamProvider
+from routstr.upstream.rate_limit import (
+    UPSTREAM_RATE_LIMIT,
+    RateLimitInfo,
+    classify_rate_limit,
+)
+
+# The exact scenario from the issue, with a realistic (fake) org identifier.
+RAW_ORG_ID = "org-abc123XYZ456def"
+RATE_LIMIT_MESSAGE = (
+    f"Rate limit reached for gpt-5.5-2026-04-23 (for limit gpt-5.5) in "
+    f"organization {RAW_ORG_ID} on tokens per min (TPM): Limit 180000000, "
+    f"Used 180000000, Requested 8929. Please try again in 2ms. Visit "
+    f"https://platform.openai.com/account/rate-limits to learn more."
+)
+
+
+def _make_request(request_id: str = "req-123") -> Mock:
+    request = Mock(spec=["method", "state"])
+    request.method = "POST"
+    request.state = Mock()
+    request.state.request_id = request_id
+    return request
+
+
+def _make_upstream_response(
+    *,
+    body: bytes,
+    status_code: int = 429,
+    content_type: str | None = "application/json",
+    extra_headers: dict[str, str] | None = None,
+) -> httpx.Response:
+    headers: dict[str, str] = {}
+    if content_type is not None:
+        headers["content-type"] = content_type
+    if extra_headers:
+        headers.update(extra_headers)
+    return httpx.Response(status_code=status_code, headers=headers, content=body)
+
+
+@pytest.fixture
+def provider() -> BaseUpstreamProvider:
+    return BaseUpstreamProvider(
+        base_url="https://privateprovider.xyz", api_key="k", provider_fee=1.0
+    )
+
+
+# --------------------------------------------------------------------------- #
+# Redaction
+# --------------------------------------------------------------------------- #
+
+
+def test_redact_org_ids_replaces_identifier() -> None:
+    assert RAW_ORG_ID not in redact_org_ids(RATE_LIMIT_MESSAGE)
+    assert "org-[REDACTED]" in redact_org_ids(RATE_LIMIT_MESSAGE)
+
+
+def test_redact_org_ids_is_idempotent() -> None:
+    once = redact_org_ids(RATE_LIMIT_MESSAGE)
+    assert redact_org_ids(once) == once
+
+
+def test_redact_org_ids_leaves_unrelated_text() -> None:
+    assert redact_org_ids("organize the org-chart") == "organize the org-chart"
+    assert redact_org_ids("") == ""
+
+
+# --------------------------------------------------------------------------- #
+# Classification
+# --------------------------------------------------------------------------- #
+
+
+def test_classify_exact_scenario() -> None:
+    info = classify_rate_limit(429, RATE_LIMIT_MESSAGE)
+    assert isinstance(info, RateLimitInfo)
+    assert info.code == UPSTREAM_RATE_LIMIT
+    assert info.model == "gpt-5.5-2026-04-23"
+    assert info.limit_name == "gpt-5.5"
+    assert info.metric == "tokens per min (TPM)"
+    assert info.limit == 180000000
+    assert info.used == 180000000
+    assert info.requested == 8929
+    assert info.retry_after_seconds == pytest.approx(0.002)
+    # Redaction-safe: no raw org id survives into the structured view.
+    assert RAW_ORG_ID not in info.message
+    assert RAW_ORG_ID not in json.dumps(info.as_details())
+
+
+def test_classify_by_status_code_without_marker() -> None:
+    info = classify_rate_limit(429, "slow down")
+    assert info is not None
+    assert info.code == UPSTREAM_RATE_LIMIT
+
+
+def test_classify_by_message_marker_without_429() -> None:
+    info = classify_rate_limit(400, "rate_limit_exceeded for this key")
+    assert info is not None
+
+
+def test_retry_after_header_takes_precedence() -> None:
+    info = classify_rate_limit(429, RATE_LIMIT_MESSAGE, {"Retry-After": "12"})
+    assert info is not None
+    assert info.retry_after_seconds == pytest.approx(12.0)
+
+
+def test_non_rate_limit_error_is_not_classified() -> None:
+    assert classify_rate_limit(400, "invalid request: missing field 'model'") is None
+    assert classify_rate_limit(500, "internal server error") is None
+
+
+# --------------------------------------------------------------------------- #
+# forward_upstream_error_response integration
+# --------------------------------------------------------------------------- #
+
+
+@pytest.mark.asyncio
+async def test_json_rate_limit_body_is_redacted_and_forwarded(
+    provider: BaseUpstreamProvider,
+) -> None:
+    body = json.dumps(
+        {"error": {"message": RATE_LIMIT_MESSAGE, "type": "rate_limit_exceeded"}}
+    ).encode()
+    upstream = _make_upstream_response(body=body, status_code=429)
+
+    response = await provider.forward_upstream_error_response(
+        _make_request(), "v1/chat/completions", upstream
+    )
+
+    assert response.status_code == 429
+    raw = bytes(response.body).decode()
+    # No raw organization id may survive in the forwarded body.
+    assert RAW_ORG_ID not in raw
+    assert "org-[REDACTED]" in raw
+    # Body remains valid JSON; the original type is preserved while a stable
+    # rate-limit code is injected so callers can switch on it.
+    payload: dict[str, Any] = json.loads(raw)
+    assert payload["error"]["type"] == "rate_limit_exceeded"
+    assert payload["error"]["code"] == UPSTREAM_RATE_LIMIT
+    assert payload["error"]["details"]["model"] == "gpt-5.5-2026-04-23"
+    # A retry hint extracted from the message is surfaced as a header.
+    assert "retry-after" in {k.lower() for k in response.headers}
+
+
+@pytest.mark.asyncio
+async def test_non_json_rate_limit_envelope_uses_stable_code(
+    provider: BaseUpstreamProvider,
+) -> None:
+    upstream = _make_upstream_response(
+        body=RATE_LIMIT_MESSAGE.encode(),
+        status_code=429,
+        content_type="text/plain",
+    )
+
+    response = await provider.forward_upstream_error_response(
+        _make_request(), "v1/chat/completions", upstream
+    )
+
+    payload: dict[str, Any] = json.loads(bytes(response.body))
+    assert payload["error"]["code"] == UPSTREAM_RATE_LIMIT
+    assert payload["error"]["details"]["model"] == "gpt-5.5-2026-04-23"
+    serialized = json.dumps(payload)
+    assert RAW_ORG_ID not in serialized
+    assert "org-[REDACTED]" in serialized
+
+
+@pytest.mark.asyncio
+async def test_non_rate_limit_json_error_unchanged(
+    provider: BaseUpstreamProvider,
+) -> None:
+    body = json.dumps(
+        {"error": {"message": "missing field 'model'", "type": "invalid_request"}}
+    ).encode()
+    upstream = _make_upstream_response(body=body, status_code=400)
+
+    response = await provider.forward_upstream_error_response(
+        _make_request(), "v1/chat/completions", upstream
+    )
+
+    assert response.status_code == 400
+    payload: dict[str, Any] = json.loads(bytes(response.body))
+    assert payload["error"]["type"] == "invalid_request"
+    assert "retry-after" not in {k.lower() for k in response.headers}
+
+
+# --------------------------------------------------------------------------- #
+# UpstreamError -> proxy response (preserves code/details/status)
+# --------------------------------------------------------------------------- #
+
+
+def test_create_upstream_error_response_preserves_structure() -> None:
+    from routstr.core.exceptions import UpstreamError
+    from routstr.payment.helpers import create_upstream_error_response
+
+    info = classify_rate_limit(429, RATE_LIMIT_MESSAGE)
+    assert info is not None
+    err = UpstreamError(
+        f"Upstream error via litellm: {RATE_LIMIT_MESSAGE}",
+        status_code=429,
+        code=info.code,
+        details=info.as_details(),
+    )
+
+    response = create_upstream_error_response(err, _make_request())
+
+    # Original upstream status is preserved (not flattened to 502).
+    assert response.status_code == 429
+    payload: dict[str, Any] = json.loads(bytes(response.body))
+    assert payload["error"]["type"] == "upstream_error"
+    assert payload["error"]["code"] == UPSTREAM_RATE_LIMIT
+    assert payload["error"]["details"]["requested"] == 8929
+    serialized = json.dumps(payload)
+    assert RAW_ORG_ID not in serialized
+    assert "org-[REDACTED]" in serialized
+
+
+def test_generic_upstream_error_still_defaults_to_502() -> None:
+    from routstr.core.exceptions import UpstreamError
+    from routstr.payment.helpers import create_upstream_error_response
+
+    err = UpstreamError("connection refused")  # status_code defaults to 502
+
+    response = create_upstream_error_response(err, _make_request())
+
+    assert response.status_code == 502
+    payload: dict[str, Any] = json.loads(bytes(response.body))
+    assert payload["error"]["type"] == "upstream_error"
+    assert payload["error"]["code"] == 502
+    assert "details" not in payload["error"]
+
+
+# --------------------------------------------------------------------------- #
+# Structured log-extra redaction
+# --------------------------------------------------------------------------- #
+
+
+def test_security_filter_redacts_org_id_in_extra() -> None:
+    import logging
+
+    from routstr.core.logging import SecurityFilter
+
+    record = logging.LogRecord(
+        name="test",
+        level=logging.ERROR,
+        pathname=__file__,
+        lineno=1,
+        msg="upstream failed",
+        args=(),
+        exc_info=None,
+    )
+    # Simulate an ``extra={"body_preview": ...}`` field carrying an org id.
+    setattr(record, "body_preview", RATE_LIMIT_MESSAGE)
+
+    assert SecurityFilter().filter(record) is True
+    redacted: str = getattr(record, "body_preview")
+    assert RAW_ORG_ID not in redacted
+    assert "org-[REDACTED]" in redacted
+
+
+def test_security_filter_redacts_org_id_in_nested_extra() -> None:
+    import logging
+
+    from routstr.core.logging import SecurityFilter
+
+    record = logging.LogRecord(
+        name="test",
+        level=logging.ERROR,
+        pathname=__file__,
+        lineno=1,
+        msg="upstream failed",
+        args=(),
+        exc_info=None,
+    )
+    # Nested structures: dict containing a list containing the org id.
+    setattr(record, "body", {"error": {"messages": [RATE_LIMIT_MESSAGE]}})
+
+    assert SecurityFilter().filter(record) is True
+    serialized = json.dumps(getattr(record, "body"))
+    assert RAW_ORG_ID not in serialized
+    assert "org-[REDACTED]" in serialized
+
+
+# --------------------------------------------------------------------------- #
+# 5xx-wrapped rate limit through forward_upstream_error_response
+# --------------------------------------------------------------------------- #
+
+
+@pytest.mark.asyncio
+async def test_5xx_wrapped_rate_limit_is_classified(
+    provider: BaseUpstreamProvider,
+) -> None:
+    # Some providers wrap a rate-limit in a 5xx envelope; classification must
+    # key off the message marker, not only the 429 status.
+    body = json.dumps({"error": {"message": RATE_LIMIT_MESSAGE}}).encode()
+    upstream = _make_upstream_response(body=body, status_code=500)
+
+    response = await provider.forward_upstream_error_response(
+        _make_request(), "v1/chat/completions", upstream
+    )
+
+    assert response.status_code == 500
+    payload: dict[str, Any] = json.loads(bytes(response.body))
+    assert payload["error"]["code"] == UPSTREAM_RATE_LIMIT
+    serialized = json.dumps(payload)
+    assert RAW_ORG_ID not in serialized
+    assert "org-[REDACTED]" in serialized
+
+
+# --------------------------------------------------------------------------- #
+# Real proxy loop: structured error surfaced + reservation reverted once
+# --------------------------------------------------------------------------- #
+
+
+@pytest.mark.asyncio
+async def test_proxy_loop_surfaces_rate_limit_and_reverts_once() -> None:
+    from routstr import proxy as proxy_module
+    from routstr.core.db import ApiKey
+    from routstr.core.exceptions import UpstreamError
+
+    info = classify_rate_limit(429, RATE_LIMIT_MESSAGE)
+    assert info is not None
+
+    key = ApiKey(hashed_key="rlkey", balance=10_000)
+
+    request = MagicMock()
+    request.method = "POST"
+    request.headers = {"authorization": "Bearer sk-rlkey"}
+    request.body = AsyncMock(return_value=b'{"model": "test-model"}')
+    request.state = MagicMock()
+    request.state.request_id = "req-rl"
+
+    upstream = MagicMock()
+    upstream.provider_type = "test"
+    upstream.prepare_headers = MagicMock(side_effect=lambda h: h)
+    upstream.forward_request = AsyncMock(
+        side_effect=UpstreamError(
+            f"Upstream error via litellm: {RATE_LIMIT_MESSAGE}",
+            status_code=429,
+            code=info.code,
+            details=info.as_details(),
+        )
+    )
+
+    session = MagicMock()
+    revert_mock = AsyncMock(return_value=True)
+
+    with (
+        patch.object(proxy_module, "get_model_instance", return_value=MagicMock()),
+        patch.object(proxy_module, "get_provider_for_model", return_value=[upstream]),
+        patch.object(
+            proxy_module, "get_max_cost_for_model", AsyncMock(return_value=1_000)
+        ),
+        patch.object(
+            proxy_module,
+            "calculate_discounted_max_cost",
+            AsyncMock(return_value=1_000),
+        ),
+        patch.object(proxy_module, "check_token_balance", MagicMock()),
+        patch.object(
+            proxy_module, "get_bearer_token_key", AsyncMock(return_value=key)
+        ),
+        patch.object(proxy_module, "pay_for_request", AsyncMock(return_value=1_000)),
+        patch.object(proxy_module, "revert_pay_for_request", revert_mock),
+    ):
+        response = await proxy_module.proxy(
+            request, "v1/chat/completions", session=session
+        )
+
+    # Original 429 status and the stable code/details survive to the client.
+    assert response.status_code == 429
+    payload: dict[str, Any] = json.loads(bytes(response.body))
+    assert payload["error"]["type"] == "upstream_error"
+    assert payload["error"]["code"] == UPSTREAM_RATE_LIMIT
+    assert payload["error"]["details"]["requested"] == 8929
+    serialized = json.dumps(payload)
+    assert RAW_ORG_ID not in serialized
+    assert "org-[REDACTED]" in serialized
+    # Single upstream failed -> reservation reverted exactly once (no double-charge).
+    revert_mock.assert_awaited_once_with(key, session, 1_000)

tests/unit/test_wallet.py

@@ -150,6 +150,7 @@ async def test_credit_balance() -> None:
     mock_key.balance = 5000000
     mock_key.hashed_key = "test_hash"
     mock_session = AsyncMock()
+    mock_session.exec.return_value.rowcount = 1
 
     # Mock session.refresh to update the balance (simulates DB reload)
     async def mock_refresh(key: ApiKey) -> None:
@@ -206,9 +207,43 @@ async def test_credit_balance_rejects_zero_amount() -> None:
     assert not mock_session.commit.called
 
 
+@pytest.mark.asyncio
+async def test_credit_balance_rejects_missing_key() -> None:
+    """A top-up must fail if the key was pruned after redemption."""
+    token_data = {
+        "token": [{"mint": "http://mint:3338", "proofs": [{"amount": 1000}]}],
+        "unit": "sat",
+    }
+    token_json = json.dumps(token_data)
+    token_b64 = base64.urlsafe_b64encode(token_json.encode()).decode()
+    token_str = f"cashuA{token_b64}"
+
+    mock_key = Mock()
+    mock_key.balance = 0
+    mock_key.hashed_key = "test_hash"
+    mock_session = AsyncMock()
+    mock_session.exec.return_value.rowcount = 0
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "cashu_mints", ["http://mint:3338"]):
+        with patch(
+            "routstr.wallet.recieve_token",
+            return_value=(1000, "sat", "http://mint:3338"),
+        ):
+            with pytest.raises(ValueError, match="disappeared"):
+                await credit_balance(token_str, mock_key, mock_session)
+
+    # UPDATE matched nothing; committing would hide the failed credit.
+    assert mock_session.exec.called
+    assert not mock_session.commit.called
+
+
 @pytest.mark.asyncio
 async def test_swap_to_primary_mint_insufficient_for_fees() -> None:
-    """Token amount is less than melt_quote.amount + melt_quote.fee_reserve."""
+    """Token amount is less than melt_quote.amount + melt_quote.fee_reserve.
+    The quote mocks are static, so every retry observes the same shortfall —
+    the swap must still give up and raise."""
     from routstr.wallet import swap_to_primary_mint
 
     mock_token = Mock()
@@ -250,50 +285,6 @@ async def test_swap_to_primary_mint_insufficient_for_fees() -> None:
     mock_token_wallet.melt.assert_not_called()
 
 
-@pytest.mark.asyncio
-async def test_swap_to_primary_mint_melt_error_wrapped() -> None:
-    """Melt failure from cashu lib is wrapped as ValueError."""
-    from routstr.wallet import swap_to_primary_mint
-
-    mock_token = Mock()
-    mock_token.mint = "http://foreign:3338"
-    mock_token.unit = "sat"
-    mock_token.amount = 5000
-    mock_token.keysets = ["keyset1"]
-    mock_token.proofs = [{"amount": 5000}]
-
-    mock_token_wallet = Mock()
-    mock_token_wallet.load_mint = AsyncMock()
-    mock_token_wallet.load_proofs = AsyncMock()
-    mock_token_wallet.get_fees_for_proofs = Mock(return_value=0)
-
-    mock_primary_wallet = Mock()
-    mock_primary_wallet.load_mint = AsyncMock()
-    mock_primary_wallet.load_proofs = AsyncMock()
-
-    mock_mint_quote = Mock()
-    mock_mint_quote.quote = "mint_quote_456"
-    mock_mint_quote.request = "lnbc1..."
-    mock_primary_wallet.request_mint = AsyncMock(return_value=mock_mint_quote)
-
-    mock_melt_quote = Mock()
-    mock_melt_quote.quote = "melt_quote_456"
-    mock_melt_quote.amount = 4940
-    mock_melt_quote.fee_reserve = 50  # total 4990 < 5000, passes fee check
-    mock_token_wallet.melt_quote = AsyncMock(return_value=mock_melt_quote)
-    mock_token_wallet.melt = AsyncMock(
-        side_effect=Exception("Provided: 5000, needed: 5100 (Code: 11000)")
-    )
-
-    from routstr.core.settings import settings
-
-    with patch.object(settings, "primary_mint", "http://primary:3338"):
-        with patch.object(settings, "primary_mint_unit", "sat"):
-            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
-                with pytest.raises(ValueError, match="Failed to melt token"):
-                    await swap_to_primary_mint(mock_token, mock_token_wallet)
-
-
 @pytest.mark.asyncio
 async def test_recieve_token_untrusted_mint() -> None:
     mock_wallet = Mock()
@@ -361,54 +352,80 @@ async def test_swap_to_primary_mint_already_on_primary() -> None:
     mock_token_wallet.melt_quote.assert_not_called()
 
 
-async def test_swap_to_primary_mint_success() -> None:
-    """Test successful swap with dynamic fee calculation."""
-    from routstr.wallet import swap_to_primary_mint
+# ---------------------------------------------------------------------------
+# Swap fee estimation and reactive retry
+#
+# Spec: the estimation pass subtracts only observed fees (no safety buffer).
+# swap_to_primary_mint then runs the mint-quote/melt-quote/melt cycle and, when
+# the foreign mint demands more than estimated (at quote or at melt time),
+# retries with the amount recomputed from the observed fee — at most 3 attempts.
+# Melt failures unrelated to fees are not retried.
+# ---------------------------------------------------------------------------
 
+
+def _make_swap_mocks(
+    token_amount: int,
+    fee_reserves: list[int],
+    input_fees: int = 0,
+    mint_url: str = "http://foreign-mint:3338",
+) -> tuple[Mock, Mock, Mock]:
+    """Return (token, token_wallet, primary_wallet) mocks that act like a mint.
+
+    Mint quotes pass the requested amount through their ``request`` field and
+    melt quotes echo that amount back, so the mocks stay consistent for
+    whatever amounts the implementation requests. ``fee_reserves`` supplies the
+    fee_reserve of each successive melt quote (the first serves the estimation
+    pass); requesting more quotes than provided fails the test.
+    """
     mock_token = Mock()
-    mock_token.mint = "http://foreign:3338"
+    mock_token.mint = mint_url
     mock_token.unit = "sat"
-    mock_token.amount = 1000
+    mock_token.amount = token_amount
     mock_token.keysets = ["keyset1"]
-    mock_token.proofs = [{"amount": 1000}]
+    mock_token.proofs = [Mock(amount=token_amount)]
 
     mock_token_wallet = Mock()
     mock_token_wallet.load_mint = AsyncMock()
     mock_token_wallet.load_proofs = AsyncMock()
-    mock_token_wallet.get_fees_for_proofs = Mock(return_value=0)
+    mock_token_wallet.get_fees_for_proofs = Mock(return_value=input_fees)
 
     mock_primary_wallet = Mock()
     mock_primary_wallet.load_mint = AsyncMock()
     mock_primary_wallet.load_proofs = AsyncMock()
+    mock_primary_wallet.available_balance = Mock(amount=0)
+    mock_primary_wallet.mint = AsyncMock(return_value=Mock())
 
-    # Mocks for the estimation phase
-    # 1. request_mint(dummy_amount=1000) -> invoice_dummy
-    # 2. melt_quote(invoice_dummy) -> fee=10
+    fees = iter(fee_reserves)
 
-    # Mocks for the execution phase
-    # 3. request_mint(minted_amount=990) -> invoice_real
-    # 4. melt_quote(invoice_real) -> amount=990, fee=10
-    # 5. melt() -> success
-    # 6. mint() -> success
+    def _next_fee() -> int:
+        try:
+            return next(fees)
+        except StopIteration:
+            raise AssertionError(
+                "more melt quotes requested than fee_reserves provided"
+            ) from None
 
-    mock_mint_quote_dummy = Mock(quote="dummy_quote", request="lnbc_dummy")
-    mock_mint_quote_real = Mock(quote="real_quote", request="lnbc_real")
-
-    # side_effect for request_mint to return dummy then real
     mock_primary_wallet.request_mint = AsyncMock(
-        side_effect=[mock_mint_quote_dummy, mock_mint_quote_real]
+        side_effect=lambda amount: Mock(quote=f"mint_quote_{amount}", request=amount)
     )
-
-    mock_melt_quote_dummy = Mock(amount=1000, fee_reserve=10)
-    mock_melt_quote_real = Mock(amount=990, fee_reserve=10)
-
-    # side_effect for melt_quote
     mock_token_wallet.melt_quote = AsyncMock(
-        side_effect=[mock_melt_quote_dummy, mock_melt_quote_real]
+        side_effect=lambda invoice: Mock(
+            quote=f"melt_quote_{invoice}", amount=invoice, fee_reserve=_next_fee()
+        )
     )
+    mock_token_wallet.melt = AsyncMock(return_value=Mock())
 
-    mock_token_wallet.melt = AsyncMock(return_value="melted_proofs")
-    mock_primary_wallet.mint = AsyncMock(return_value="minted_proofs")
+    return mock_token, mock_token_wallet, mock_primary_wallet
+
+
+@pytest.mark.asyncio
+async def test_swap_to_primary_mint_success() -> None:
+    """No retry needed: real quote matches the estimate, full net amount minted."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[10, 10]
+    )
 
     from routstr.core.settings import settings
 
@@ -419,17 +436,659 @@ async def test_swap_to_primary_mint_success() -> None:
                     mock_token, mock_token_wallet
                 )
 
-                assert amount == 990  # 1000 - 10
-                assert unit == "sat"
-                assert mint == "http://primary:3338"
+    assert amount == 990  # 1000 - fee_reserve(10), no buffer subtracted
+    assert unit == "sat"
+    assert mint == "http://primary:3338"
+    assert mock_primary_wallet.request_mint.call_count == 2
+    mock_primary_wallet.request_mint.assert_any_call(1000)
+    mock_primary_wallet.request_mint.assert_any_call(990)
+    assert mock_token_wallet.melt_quote.call_count == 2
+    assert mock_token_wallet.melt.call_count == 1
+    assert mock_primary_wallet.mint.called
 
-                # Verify call order/counts
-                assert mock_primary_wallet.request_mint.call_count == 2
-                # First call with full amount for estimation
-                mock_primary_wallet.request_mint.assert_any_call(1000)
-                # Second call with calculated amount
-                mock_primary_wallet.request_mint.assert_any_call(990)
 
-                assert mock_token_wallet.melt_quote.call_count == 2
-                assert mock_token_wallet.melt.called
-                assert mock_primary_wallet.mint.called
+@pytest.mark.asyncio
+@pytest.mark.parametrize("fee_reserve", [1, 10, 100])
+async def test_calculate_swap_amount_subtracts_only_observed_fees(
+    fee_reserve: int,
+) -> None:
+    """Estimation: minted_amount = token - fee_reserve, with no safety buffer."""
+    from routstr.wallet import _calculate_swap_amount
+
+    _, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[fee_reserve]
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            result = await _calculate_swap_amount(
+                amount_msat=1_000_000,
+                token_unit="sat",
+                token_mint_url="http://foreign-mint:3338",
+                token_wallet=mock_token_wallet,
+                primary_wallet=mock_primary_wallet,
+                proofs=[],
+            )
+
+    assert result == 1000 - fee_reserve
+
+
+@pytest.mark.asyncio
+async def test_calculate_swap_amount_includes_input_fees() -> None:
+    """Estimation subtracts NUT-02 input fees alongside the melt fee_reserve."""
+    from routstr.wallet import _calculate_swap_amount
+
+    _, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        500, fee_reserves=[10], input_fees=3
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            result = await _calculate_swap_amount(
+                amount_msat=500_000,
+                token_unit="sat",
+                token_mint_url="http://foreign-mint:3338",
+                token_wallet=mock_token_wallet,
+                primary_wallet=mock_primary_wallet,
+                proofs=[],
+            )
+
+    assert result == 487  # 500 - 10 - 3
+
+
+@pytest.mark.asyncio
+async def test_swap_retries_when_real_quote_exceeds_estimate() -> None:
+    """The real melt quote demands a higher fee than the estimate (20 → 23).
+    Instead of failing, the swap recomputes the amount from the observed fee
+    and re-quotes: 1000 - 23 = 977, which fits (977 + 23 <= 1000)."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[20, 23, 23]
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                amount, unit, mint = await swap_to_primary_mint(
+                    mock_token, mock_token_wallet
+                )
+
+    assert amount == 977
+    assert unit == "sat"
+    mock_primary_wallet.request_mint.assert_any_call(980)
+    mock_primary_wallet.request_mint.assert_any_call(977)
+    assert mock_token_wallet.melt_quote.call_count == 3  # estimation + 2 attempts
+    assert mock_token_wallet.melt.call_count == 1
+
+
+@pytest.mark.asyncio
+async def test_swap_retries_when_melt_demands_more_than_quoted() -> None:
+    """The mint.cubabitcoin.org incident: every quote reports fee_reserve=1,
+    but the mint demands 2 sats at melt time ("Provided: 179, needed: 180").
+    The swap must retry with a smaller invoice (177) so the second melt fits,
+    instead of failing the topup."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        179, fee_reserves=[1, 1, 1], mint_url="http://mint.cubabitcoin.org"
+    )
+    mock_token_wallet.melt.side_effect = [
+        Exception(
+            "Mint Error: not enough inputs provided for melt. "
+            "Provided: 179, needed: 180 (Code: 11000)"
+        ),
+        Mock(),
+    ]
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                amount, unit, mint = await swap_to_primary_mint(
+                    mock_token, mock_token_wallet
+                )
+
+    assert amount == 177  # 179 - 1 (estimate) - 1 (observed melt shortfall)
+    assert mock_token_wallet.melt.call_count == 2
+    mock_primary_wallet.request_mint.assert_any_call(178)
+    mock_primary_wallet.request_mint.assert_any_call(177)
+
+
+@pytest.mark.asyncio
+async def test_swap_retries_on_cdk_unbalanced_error() -> None:
+    """cdk-based mints report insufficient melt inputs as the registered code
+    11005 (TransactionUnbalanced) with their own message wording — no
+    Provided/needed amounts to parse. The retry must classify it by code and
+    fall back to shrinking by 1."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        179, fee_reserves=[1, 1, 1]
+    )
+    mock_token_wallet.melt.side_effect = [
+        Exception("Mint Error: Transaction unbalanced: 179, 178, 2 (Code: 11005)"),
+        Mock(),
+    ]
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                amount, unit, mint = await swap_to_primary_mint(
+                    mock_token, mock_token_wallet
+                )
+
+    assert amount == 177
+    assert mock_token_wallet.melt.call_count == 2
+
+
+@pytest.mark.asyncio
+async def test_swap_quote_retries_exhausted() -> None:
+    """A mint that escalates fee_reserve on every re-quote exhausts the retry
+    budget (3 attempts) and fails cleanly; melt is never executed."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[1, 10, 25, 50]
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                with pytest.raises(ValueError, match="insufficient to cover melt fees"):
+                    await swap_to_primary_mint(mock_token, mock_token_wallet)
+
+    assert mock_token_wallet.melt_quote.call_count == 4  # estimation + 3 attempts
+    mock_token_wallet.melt.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_swap_melt_retries_exhausted() -> None:
+    """A mint that always demands more at melt time than it quoted exhausts
+    the retry budget; the last melt failure is wrapped as ValueError."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        5000, fee_reserves=[50, 50, 50, 50]
+    )
+    mock_token_wallet.melt = AsyncMock(
+        side_effect=Exception(
+            "Mint Error: not enough inputs provided for melt. "
+            "Provided: 5000, needed: 5200 (Code: 11000)"
+        )
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                with pytest.raises(ValueError, match="Failed to melt token"):
+                    await swap_to_primary_mint(mock_token, mock_token_wallet)
+
+    assert mock_token_wallet.melt.call_count == 3
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "primary_unit,token_unit,amount_msat,fees,expected",
+    [
+        ("sat", "sat", 179_000, 2, 177),
+        ("msat", "sat", 179_000, 2, 177_000),
+        ("sat", "msat", 179_000, 2_000, 177),
+    ],
+)
+async def test_net_minted_amount_unit_conversions(
+    primary_unit: str, token_unit: str, amount_msat: int, fees: int, expected: int
+) -> None:
+    """Fee subtraction converts correctly between sat and msat on either side."""
+    from routstr.core.settings import settings
+    from routstr.wallet import _net_minted_amount
+
+    with patch.object(settings, "primary_mint_unit", primary_unit):
+        assert _net_minted_amount(amount_msat, token_unit, fees) == expected
+
+
+@pytest.mark.parametrize(
+    "message,expected",
+    [
+        # nutshell: retryable with exact shortfall from the detail text
+        (
+            "Mint Error: not enough inputs provided for melt. "
+            "Provided: 179, needed: 182 (Code: 11000)",
+            3,
+        ),
+        # verbatim production error from issue #468, including cashu-py's
+        # "could not pay invoice" wrapper around the mint detail
+        (
+            "could not pay invoice: Mint Error: not enough inputs provided "
+            "for melt. Provided: 179, needed: 180 (Code: 11000)",
+            1,
+        ),
+        # cdk: registered TransactionUnbalanced code, no parsable amounts
+        ("Mint Error: Transaction unbalanced: 179, 178, 2 (Code: 11005)", 1),
+        # nutshell wording without a code suffix
+        ("not enough inputs provided for melt", 1),
+        # nonsensical amounts (needed <= provided) fall back to the minimal step
+        (
+            "Mint Error: not enough inputs provided for melt. "
+            "Provided: 180, needed: 179 (Code: 11000)",
+            1,
+        ),
+        # a generic 11000 without the shortfall text is not a fee shortfall:
+        # 11000 is nutshell's catch-all TransactionError, so retrying (shrinking
+        # the invoice) would never help and only masks the real error
+        ("Mint Error: Duplicate inputs provided. (Code: 11000)", None),
+        # spent proofs must never be retried: the funds are gone
+        ("Mint Error: Token already spent. (Code: 11001)", None),
+        # Lightning failures must never be retried: a smaller invoice won't help
+        ("Mint Error: Lightning payment failed. (Code: 20004)", None),
+        # unrecognizable errors (timeouts, bugs) must never be retried
+        ("Connection timeout", None),
+    ],
+)
+def test_melt_shortfall_classifier(message: str, expected: int | None) -> None:
+    """Retry classification across mint implementations and failure classes."""
+    from routstr.wallet import _melt_insufficient_shortfall
+
+    assert _melt_insufficient_shortfall(Exception(message)) == expected
+
+
+@pytest.mark.asyncio
+async def test_calculate_swap_amount_same_mint_short_circuit() -> None:
+    """When the token is already on the primary mint no fees apply and no
+    quotes are requested."""
+    from routstr.wallet import _calculate_swap_amount
+
+    _, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[]
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            result = await _calculate_swap_amount(
+                amount_msat=1_000_000,
+                token_unit="sat",
+                token_mint_url="http://primary:3338",
+                token_wallet=mock_token_wallet,
+                primary_wallet=mock_primary_wallet,
+                proofs=[],
+            )
+
+    assert result == 1000
+    mock_primary_wallet.request_mint.assert_not_called()
+    mock_token_wallet.melt_quote.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_calculate_swap_amount_msat_primary_unit() -> None:
+    """With an msat primary mint the dummy quote and result stay in msats."""
+    from routstr.wallet import _calculate_swap_amount
+
+    _, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        179, fee_reserves=[2]
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "msat"):
+            result = await _calculate_swap_amount(
+                amount_msat=179_000,
+                token_unit="sat",
+                token_mint_url="http://foreign-mint:3338",
+                token_wallet=mock_token_wallet,
+                primary_wallet=mock_primary_wallet,
+                proofs=[],
+            )
+
+    assert result == 177_000  # 179_000 msat - 2 sat fee
+    mock_primary_wallet.request_mint.assert_called_once_with(179_000)
+
+
+@pytest.mark.asyncio
+async def test_calculate_swap_amount_fees_exceed_token() -> None:
+    """Fees larger than the token itself fail fast, before any melt."""
+    from routstr.wallet import _calculate_swap_amount
+
+    _, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        179, fee_reserves=[200]
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with pytest.raises(ValueError, match="exceed token amount"):
+                await _calculate_swap_amount(
+                    amount_msat=179_000,
+                    token_unit="sat",
+                    token_mint_url="http://foreign-mint:3338",
+                    token_wallet=mock_token_wallet,
+                    primary_wallet=mock_primary_wallet,
+                    proofs=[],
+                )
+
+
+@pytest.mark.asyncio
+async def test_calculate_swap_amount_wraps_estimation_failure() -> None:
+    """Estimation infrastructure failures surface as a single clear ValueError."""
+    from routstr.wallet import _calculate_swap_amount
+
+    _, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        179, fee_reserves=[]
+    )
+    mock_primary_wallet.request_mint = AsyncMock(
+        side_effect=Exception("mint offline")
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with pytest.raises(ValueError, match="Failed to estimate fees"):
+                await _calculate_swap_amount(
+                    amount_msat=179_000,
+                    token_unit="sat",
+                    token_mint_url="http://foreign-mint:3338",
+                    token_wallet=mock_token_wallet,
+                    primary_wallet=mock_primary_wallet,
+                    proofs=[],
+                )
+
+
+@pytest.mark.asyncio
+async def test_swap_coerces_non_integer_amount() -> None:
+    """Token amounts arriving as floats are coerced before any arithmetic."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[10, 10]
+    )
+    mock_token.amount = 1000.0
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                amount, unit, mint = await swap_to_primary_mint(
+                    mock_token, mock_token_wallet
+                )
+
+    assert amount == 990
+    assert isinstance(amount, int)
+
+
+@pytest.mark.asyncio
+async def test_swap_rejects_unknown_unit() -> None:
+    """Units other than sat/msat are rejected before any quote is requested."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[]
+    )
+    mock_token.unit = "usd"
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                with pytest.raises(ValueError, match="Invalid unit"):
+                    await swap_to_primary_mint(mock_token, mock_token_wallet)
+
+    mock_primary_wallet.request_mint.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_swap_msat_token_already_on_primary() -> None:
+    """msat-denominated tokens on the primary mint short-circuit unchanged."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, _ = _make_swap_mocks(
+        179_000, fee_reserves=[], mint_url="http://primary:3338"
+    )
+    mock_token.unit = "msat"
+    mock_token_wallet.split = AsyncMock()
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_token_wallet):
+                amount, unit, mint = await swap_to_primary_mint(
+                    mock_token, mock_token_wallet
+                )
+
+    assert (amount, unit, mint) == (179_000, "msat", "http://primary:3338")
+
+
+# ---------------------------------------------------------------------------
+# Mint-on-primary failure handling after a successful melt
+#
+# At this point the foreign proofs are already spent: failures here mean funds
+# are in limbo, so errors must propagate (never be swallowed) and recovery must
+# never credit proofs the wallet does not actually hold.
+# ---------------------------------------------------------------------------
+
+
+def _with_recovery_mocks(
+    mock_primary_wallet: Mock, mint_error: str, balances: list[int]
+) -> None:
+    """Make primary mint() fail and stage available_balance per load_proofs call."""
+    mock_primary_wallet.mint = AsyncMock(side_effect=Exception(mint_error))
+    mock_primary_wallet.keysets = ["keyset_primary"]
+    balance_iter = iter(balances)
+
+    def advance_balance(reload: bool = False) -> None:
+        mock_primary_wallet.available_balance = Mock(amount=next(balance_iter))
+
+    mock_primary_wallet.load_proofs = AsyncMock(side_effect=advance_balance)
+    mock_primary_wallet.restore_tokens_for_keyset = AsyncMock()
+
+
+@pytest.mark.asyncio
+async def test_swap_mint_failure_propagates_unwrapped() -> None:
+    """A non-recoverable mint failure after melt propagates as-is (a 500, not a
+    client error): the melt already spent the foreign proofs."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[10, 10]
+    )
+    _with_recovery_mocks(
+        mock_primary_wallet, "Mint Error: Quote is expired (Code: 20007)", [0]
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                with pytest.raises(Exception, match="Quote is expired") as exc_info:
+                    await swap_to_primary_mint(mock_token, mock_token_wallet)
+
+    assert type(exc_info.value) is Exception  # original error, not wrapped
+    assert mock_token_wallet.melt.call_count == 1
+    mock_primary_wallet.restore_tokens_for_keyset.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_swap_recovers_orphaned_proofs_on_outputs_already_signed() -> None:
+    """11003 (outputs already signed): a recovery scan that restores the full
+    minted amount lets the swap complete normally."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[10, 10]
+    )
+    _with_recovery_mocks(
+        mock_primary_wallet,
+        "Mint Error: outputs already signed (Code: 11003)",
+        [0, 990],  # pre-mint balance, post-recovery balance
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                amount, unit, mint = await swap_to_primary_mint(
+                    mock_token, mock_token_wallet
+                )
+
+    assert amount == 990
+    mock_primary_wallet.restore_tokens_for_keyset.assert_awaited_once_with(
+        "keyset_primary", to=1, batch=25
+    )
+
+
+@pytest.mark.asyncio
+async def test_swap_recovery_shortfall_refuses_credit() -> None:
+    """When the recovery scan restores less than the minted amount, the swap
+    must fail rather than credit proofs the wallet does not hold."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[10, 10]
+    )
+    _with_recovery_mocks(
+        mock_primary_wallet,
+        "Mint Error: outputs already signed (Code: 11003)",
+        [0, 100],  # recovery restores only 100 of the expected 990
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                with pytest.raises(ValueError, match="Swap recovery failed"):
+                    await swap_to_primary_mint(mock_token, mock_token_wallet)
+
+
+@pytest.mark.asyncio
+async def test_swap_recovery_failure_wrapped() -> None:
+    """When the recovery scan itself fails, the error is wrapped and raised —
+    never swallowed."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[10, 10]
+    )
+    _with_recovery_mocks(
+        mock_primary_wallet,
+        "Mint Error: outputs already signed (Code: 11003)",
+        [0],
+    )
+    mock_primary_wallet.restore_tokens_for_keyset = AsyncMock(
+        side_effect=Exception("wallet db locked")
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                with pytest.raises(ValueError, match="recovery unsuccessful"):
+                    await swap_to_primary_mint(mock_token, mock_token_wallet)
+
+
+@pytest.mark.asyncio
+async def test_recieve_token_rejects_multiple_keysets() -> None:
+    """Multi-keyset tokens are rejected before touching any wallet."""
+    with patch("routstr.wallet.deserialize_token_from_string") as mock_deserialize:
+        mock_token = Mock()
+        mock_token.keysets = ["keyset1", "keyset2"]
+        mock_deserialize.return_value = mock_token
+
+        with pytest.raises(ValueError, match="Multiple keysets"):
+            await recieve_token("cashuAmultikeyset")
+
+
+@pytest.mark.asyncio
+async def test_credit_balance_msat_unit_not_converted() -> None:
+    """msat-denominated redemptions are credited as-is, without a 1000x."""
+    mock_key = Mock()
+    mock_key.balance = 0
+    mock_key.hashed_key = "test_hash"
+    mock_session = AsyncMock()
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "cashu_mints", ["http://mint:3338"]):
+        with patch(
+            "routstr.wallet.recieve_token",
+            return_value=(1_000_000, "msat", "http://mint:3338"),
+        ):
+            amount = await credit_balance("cashuAtest", mock_key, mock_session)
+
+    assert amount == 1_000_000
+    assert mock_session.commit.called
+
+
+@pytest.mark.asyncio
+async def test_credit_balance_survives_audit_store_failure() -> None:
+    """A failure writing the CashuTransaction history record must not undo the
+    already-committed balance credit. (The silent swallow is a known
+    audit-trail gap slated for its own fix — this test pins the financial
+    invariant that the user keeps their credit, not the swallow itself.)"""
+    mock_key = Mock()
+    mock_key.balance = 0
+    mock_key.hashed_key = "test_hash"
+    mock_session = AsyncMock()
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "cashu_mints", ["http://mint:3338"]):
+        with patch(
+            "routstr.wallet.recieve_token",
+            return_value=(1000, "sat", "http://mint:3338"),
+        ):
+            with patch(
+                "routstr.wallet.store_cashu_transaction",
+                side_effect=Exception("history table locked"),
+            ):
+                amount = await credit_balance("cashuAtest", mock_key, mock_session)
+
+    assert amount == 1_000_000
+    assert mock_session.commit.called
+
+
+@pytest.mark.asyncio
+async def test_swap_does_not_retry_on_payment_failure() -> None:
+    """Melt failures unrelated to fees (e.g. routing failure) are not retried:
+    a smaller invoice would not help, and the error must surface immediately."""
+    from routstr.wallet import swap_to_primary_mint
+
+    mock_token, mock_token_wallet, mock_primary_wallet = _make_swap_mocks(
+        1000, fee_reserves=[10, 10]
+    )
+    mock_token_wallet.melt = AsyncMock(
+        side_effect=Exception("Mint Error: Lightning payment failed. (Code: 20004)")
+    )
+
+    from routstr.core.settings import settings
+
+    with patch.object(settings, "primary_mint", "http://primary:3338"):
+        with patch.object(settings, "primary_mint_unit", "sat"):
+            with patch("routstr.wallet.get_wallet", return_value=mock_primary_wallet):
+                with pytest.raises(ValueError, match="Failed to melt token"):
+                    await swap_to_primary_mint(mock_token, mock_token_wallet)
+
+    assert mock_token_wallet.melt.call_count == 1
+    assert mock_primary_wallet.request_mint.call_count == 2

ui/README.md

@@ -1,36 +1,44 @@
-This is a [Next.js](https://nextjs.org) project bootstrapped with [`create-next-app`](https://nextjs.org/docs/app/api-reference/cli/create-next-app).
+# Routstr node admin UI
 
-## Getting Started
+A [Next.js](https://nextjs.org) app (App Router, **static export**) that provides the
+admin dashboard for a `routstr-core` node: login, settings, providers, balances,
+transactions, usage, and logs.
 
-First, run the development server:
+There is no separate web server in production. `next build` produces a fully static
+export (`next.config.ts` sets `output: 'export'`), and the FastAPI backend serves it
+directly from `../ui_out/` (see `routstr/core/main.py`). So the UI and the API are
+served from the **same origin** in production.
 
-```bash
-npm run dev
-# or
-yarn dev
-# or
-pnpm dev
-# or
-bun dev
-```
+## Developing the UI (hot reload)
 
-Open [http://localhost:3000](http://localhost:3000) with your browser to see the result.
+The everyday loop runs two processes side by side — you do **not** rebuild the static
+export while developing:
 
-You can start editing the page by modifying `app/page.tsx`. The page auto-updates as you edit the file.
+1. Start the backend on `:8000` — from the repo root: `make docker-up` (or
+   `uvicorn routstr.core.main:app --reload`).
+2. Start the Next.js dev server on `:3000` — from the repo root: `make ui-dev`
+   (or `cd ui && pnpm dev`). Edits hot-reload instantly.
 
-This project uses [`next/font`](https://nextjs.org/docs/app/building-your-application/optimizing/fonts) to automatically optimize and load [Geist](https://vercel.com/font), a new font family for Vercel.
+Open http://localhost:3000. With no `NEXT_PUBLIC_API_URL` set, the UI falls back to
+`http://127.0.0.1:8000` in development (see `lib/api/services/configuration.ts`), so it
+talks to the local backend out of the box.
 
-## Learn More
+Because dev is cross-origin (`:3000` → `:8000`), it relies on the backend's CORS
+allowing the UI origin. The default `cors_origins` is `["*"]`; if you tighten CORS,
+keep `http://localhost:3000` allowed for development.
 
-To learn more about Next.js, take a look at the following resources:
+## Building the integrated/static UI (what production serves)
 
-- [Next.js Documentation](https://nextjs.org/docs) - learn about Next.js features and API.
-- [Learn Next.js](https://nextjs.org/learn) - an interactive Next.js tutorial.
+To produce the bundle that FastAPI serves from `../ui_out/`:
 
-You can check out [the Next.js GitHub repository](https://github.com/vercel/next.js) - your feedback and contributions are welcome!
+- `make ui-build` — builds with local Node/pnpm (`scripts/build-ui.sh`), then moves
+  `ui/out/*` to `../ui_out/`.
+- `make ui-build-docker` — same, but inside Docker (no local Node needed).
 
-## Deploy on Vercel
+`NEXT_PUBLIC_*` variables are read from the repo-root `.env` at build time and baked in.
+For a same-origin deployment leave `NEXT_PUBLIC_API_URL` empty (relative paths); the UI
+uses `window.location.origin` at runtime. After building, start the backend and open
+http://localhost:8000 — the dashboard is served at `/` and `/admin`.
 
-The easiest way to deploy your Next.js app is to use the [Vercel Platform](https://vercel.com/new?utm_medium=default-template&filter=next.js&utm_source=create-next-app&utm_campaign=create-next-app-readme) from the creators of Next.js.
-
-Check out our [Next.js deployment documentation](https://nextjs.org/docs/app/building-your-application/deploying) for more details.
+If `../ui_out/` does not exist, the backend logs a warning at startup and serves the API
+only (hitting a UI route returns a small JSON fallback instead of the dashboard).

ui/components/add-provider-model-dialog.tsx

@@ -68,6 +68,8 @@ const FormSchema = z.object({
   upstream_provider_id: z.string().default(''),
   input_cost: z.coerce.number().min(0).default(0),
   output_cost: z.coerce.number().min(0).default(0),
+  cache_read_cost: z.coerce.number().min(0).default(0),
+  cache_write_cost: z.coerce.number().min(0).default(0),
   request_cost: z.coerce.number().min(0).default(0),
   image_cost: z.coerce.number().min(0).default(0),
   web_search_cost: z.coerce.number().min(0).default(0),
@@ -125,6 +127,8 @@ export function AddProviderModelDialog({
       upstream_provider_id: '',
       input_cost: 0,
       output_cost: 0,
+      cache_read_cost: 0,
+      cache_write_cost: 0,
       request_cost: 0,
       image_cost: 0,
       web_search_cost: 0,
@@ -190,6 +194,8 @@ export function AddProviderModelDialog({
             : initialData.upstream_provider_id?.toString() || '',
         input_cost: pricing?.prompt ?? 0,
         output_cost: pricing?.completion ?? 0,
+        cache_read_cost: pricing?.input_cache_read ?? 0,
+        cache_write_cost: pricing?.input_cache_write ?? 0,
         request_cost: pricing?.request ?? 0,
         image_cost: pricing?.image ?? 0,
         web_search_cost: pricing?.web_search ?? 0,
@@ -231,6 +237,8 @@ export function AddProviderModelDialog({
         upstream_provider_id: '',
         input_cost: 0,
         output_cost: 0,
+        cache_read_cost: 0,
+        cache_write_cost: 0,
         request_cost: 0,
         image_cost: 0,
         web_search_cost: 0,
@@ -294,6 +302,8 @@ export function AddProviderModelDialog({
     );
     form.setValue('input_cost', pricing?.prompt ?? 0);
     form.setValue('output_cost', pricing?.completion ?? 0);
+    form.setValue('cache_read_cost', pricing?.input_cache_read ?? 0);
+    form.setValue('cache_write_cost', pricing?.input_cache_write ?? 0);
     form.setValue('request_cost', pricing?.request ?? 0);
     form.setValue('image_cost', pricing?.image ?? 0);
     form.setValue('web_search_cost', pricing?.web_search ?? 0);
@@ -365,6 +375,8 @@ export function AddProviderModelDialog({
         pricing: {
           prompt: data.input_cost,
           completion: data.output_cost,
+          input_cache_read: data.cache_read_cost,
+          input_cache_write: data.cache_write_cost,
           request: data.request_cost,
           image: data.image_cost,
           web_search: data.web_search_cost,
@@ -810,6 +822,38 @@ export function AddProviderModelDialog({
                     </FormItem>
                   )}
                 />
+                <FormField
+                  control={form.control}
+                  name='cache_read_cost'
+                  render={({ field }) => (
+                    <FormItem>
+                      <FormLabel>Cache Read Cost</FormLabel>
+                      <FormControl>
+                        <Input type='number' step='0.000001' {...field} />
+                      </FormControl>
+                      <FormDescription>
+                        Discounted cached-input read price per 1M tokens.
+                      </FormDescription>
+                      <FormMessage />
+                    </FormItem>
+                  )}
+                />
+                <FormField
+                  control={form.control}
+                  name='cache_write_cost'
+                  render={({ field }) => (
+                    <FormItem>
+                      <FormLabel>Cache Write Cost</FormLabel>
+                      <FormControl>
+                        <Input type='number' step='0.000001' {...field} />
+                      </FormControl>
+                      <FormDescription>
+                        Cached-input creation price per 1M tokens.
+                      </FormDescription>
+                      <FormMessage />
+                    </FormItem>
+                  )}
+                />
                 <FormField
                   control={form.control}
                   name='request_cost'

ui/components/provider-form-fields.tsx

@@ -118,6 +118,26 @@ export function ProviderFormFields({
         />
       )}
 
+      <div className='grid gap-2'>
+        <Label htmlFor={`${idPrefix}slug`}>
+          Slug {mode === 'create' ? '(optional, auto-generated)' : ''}
+        </Label>
+        <Input
+          id={`${idPrefix}slug`}
+          value={formData.slug || ''}
+          onChange={(e) =>
+            setFormData((prev) => ({
+              ...prev,
+              slug: e.target.value || undefined,
+            }))
+          }
+          placeholder='e.g. openai-prod'
+        />
+        <p className='text-muted-foreground text-xs'>
+          Stable external key used to update this provider via the admin API.
+        </p>
+      </div>
+
       <div className='grid gap-2'>
         <Label htmlFor={`${idPrefix}base_url`}>Base URL</Label>
         <Input

ui/lib/api/services/admin.ts

@@ -14,6 +14,7 @@ export const ProviderTypeSchema = z.object({
 
 export const UpstreamProviderSchema = z.object({
   id: z.number(),
+  slug: z.string().nullable().optional(),
   provider_type: z.string(),
   base_url: z.string(),
   api_key: z.string().optional(),
@@ -31,6 +32,7 @@ export const CreateUpstreamProviderSchema = z.object({
   enabled: z.boolean().default(true),
   provider_fee: z.number().optional(),
   provider_settings: z.record(z.string(), z.any()).nullable().optional(),
+  slug: z.string().optional(),
 });
 
 export const UpdateUpstreamProviderSchema = z.object({
@@ -41,6 +43,7 @@ export const UpdateUpstreamProviderSchema = z.object({
   enabled: z.boolean().optional(),
   provider_fee: z.number().optional(),
   provider_settings: z.record(z.string(), z.any()).nullable().optional(),
+  slug: z.string().optional(),
 });
 
 export const AdminModelPricingSchema = z.object({
@@ -50,6 +53,8 @@ export const AdminModelPricingSchema = z.object({
   image: z.number().optional(),
   web_search: z.number().optional(),
   internal_reasoning: z.number().optional(),
+  input_cache_read: z.number().optional(),
+  input_cache_write: z.number().optional(),
 });
 
 export const AdminModelArchitectureSchema = z.object({
@@ -146,7 +151,7 @@ export class AdminService {
     if (!pricing) return pricing;
     const result = { ...pricing };
 
-    // Only prompt and completion are per-token and need scaling to per-1M
+    // Token-priced fields are stored per-token by the API and shown per-1M in the UI.
     const convertField = (field: string) => {
       const val = result[field];
       if (val !== undefined && val !== null) {
@@ -161,6 +166,8 @@ export class AdminService {
 
     convertField('prompt');
     convertField('completion');
+    convertField('input_cache_read');
+    convertField('input_cache_write');
 
     // Other fields (request, image, etc.) are already flat fees (per item)
     // so we do NOT scale them.
@@ -174,7 +181,7 @@ export class AdminService {
     if (!pricing) return pricing;
     const result = { ...pricing };
 
-    // Only prompt and completion are per-1M in UI and need scaling down to per-token
+    // Token-priced fields are per-1M in the UI and need scaling down to per-token.
     const convertField = (field: string) => {
       const val = result[field];
       if (val !== undefined && val !== null) {
@@ -187,6 +194,8 @@ export class AdminService {
 
     convertField('prompt');
     convertField('completion');
+    convertField('input_cache_read');
+    convertField('input_cache_write');
 
     // Other fields stay as flat fees
 

uv.lock

@@ -17,7 +17,7 @@ wheels = [
 
 [[package]]
 name = "aiohttp"
-version = "3.12.15"
+version = "3.14.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "aiohappyeyeballs" },
@@ -26,61 +26,111 @@ dependencies = [
     { name = "frozenlist" },
     { name = "multidict" },
     { name = "propcache" },
+    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
     { name = "yarl" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/9b/e7/d92a237d8802ca88483906c388f7c201bbe96cd80a165ffd0ac2f6a8d59f/aiohttp-3.12.15.tar.gz", hash = "sha256:4fc61385e9c98d72fcdf47e6dd81833f47b2f77c114c29cd64a361be57a763a2", size = 7823716, upload-time = "2025-07-29T05:52:32.215Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/82/78/8ea7308cac6934de8c74a14f3d5f65d1c89287426688be79538d0e5c013d/aiohttp-3.14.1.tar.gz", hash = "sha256:307f2cff90a764d329e77040603fa032db89c5c24fdad50c4c15334cba744035", size = 7955794, upload-time = "2026-06-07T21:09:35.529Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/20/19/9e86722ec8e835959bd97ce8c1efa78cf361fa4531fca372551abcc9cdd6/aiohttp-3.12.15-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:d3ce17ce0220383a0f9ea07175eeaa6aa13ae5a41f30bc61d84df17f0e9b1117", size = 711246, upload-time = "2025-07-29T05:50:15.937Z" },
-    { url = "https://files.pythonhosted.org/packages/71/f9/0a31fcb1a7d4629ac9d8f01f1cb9242e2f9943f47f5d03215af91c3c1a26/aiohttp-3.12.15-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:010cc9bbd06db80fe234d9003f67e97a10fe003bfbedb40da7d71c1008eda0fe", size = 483515, upload-time = "2025-07-29T05:50:17.442Z" },
-    { url = "https://files.pythonhosted.org/packages/62/6c/94846f576f1d11df0c2e41d3001000527c0fdf63fce7e69b3927a731325d/aiohttp-3.12.15-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:3f9d7c55b41ed687b9d7165b17672340187f87a773c98236c987f08c858145a9", size = 471776, upload-time = "2025-07-29T05:50:19.568Z" },
-    { url = "https://files.pythonhosted.org/packages/f8/6c/f766d0aaafcee0447fad0328da780d344489c042e25cd58fde566bf40aed/aiohttp-3.12.15-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:bc4fbc61bb3548d3b482f9ac7ddd0f18c67e4225aaa4e8552b9f1ac7e6bda9e5", size = 1741977, upload-time = "2025-07-29T05:50:21.665Z" },
-    { url = "https://files.pythonhosted.org/packages/17/e5/fb779a05ba6ff44d7bc1e9d24c644e876bfff5abe5454f7b854cace1b9cc/aiohttp-3.12.15-cp311-cp311-manylinux_2_17_armv7l.manylinux2014_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:7fbc8a7c410bb3ad5d595bb7118147dfbb6449d862cc1125cf8867cb337e8728", size = 1690645, upload-time = "2025-07-29T05:50:23.333Z" },
-    { url = "https://files.pythonhosted.org/packages/37/4e/a22e799c2035f5d6a4ad2cf8e7c1d1bd0923192871dd6e367dafb158b14c/aiohttp-3.12.15-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:74dad41b3458dbb0511e760fb355bb0b6689e0630de8a22b1b62a98777136e16", size = 1789437, upload-time = "2025-07-29T05:50:25.007Z" },
-    { url = "https://files.pythonhosted.org/packages/28/e5/55a33b991f6433569babb56018b2fb8fb9146424f8b3a0c8ecca80556762/aiohttp-3.12.15-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:3b6f0af863cf17e6222b1735a756d664159e58855da99cfe965134a3ff63b0b0", size = 1828482, upload-time = "2025-07-29T05:50:26.693Z" },
-    { url = "https://files.pythonhosted.org/packages/c6/82/1ddf0ea4f2f3afe79dffed5e8a246737cff6cbe781887a6a170299e33204/aiohttp-3.12.15-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b5b7fe4972d48a4da367043b8e023fb70a04d1490aa7d68800e465d1b97e493b", size = 1730944, upload-time = "2025-07-29T05:50:28.382Z" },
-    { url = "https://files.pythonhosted.org/packages/1b/96/784c785674117b4cb3877522a177ba1b5e4db9ce0fd519430b5de76eec90/aiohttp-3.12.15-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:6443cca89553b7a5485331bc9bedb2342b08d073fa10b8c7d1c60579c4a7b9bd", size = 1668020, upload-time = "2025-07-29T05:50:30.032Z" },
-    { url = "https://files.pythonhosted.org/packages/12/8a/8b75f203ea7e5c21c0920d84dd24a5c0e971fe1e9b9ebbf29ae7e8e39790/aiohttp-3.12.15-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:6c5f40ec615e5264f44b4282ee27628cea221fcad52f27405b80abb346d9f3f8", size = 1716292, upload-time = "2025-07-29T05:50:31.983Z" },
-    { url = "https://files.pythonhosted.org/packages/47/0b/a1451543475bb6b86a5cfc27861e52b14085ae232896a2654ff1231c0992/aiohttp-3.12.15-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:2abbb216a1d3a2fe86dbd2edce20cdc5e9ad0be6378455b05ec7f77361b3ab50", size = 1711451, upload-time = "2025-07-29T05:50:33.989Z" },
-    { url = "https://files.pythonhosted.org/packages/55/fd/793a23a197cc2f0d29188805cfc93aa613407f07e5f9da5cd1366afd9d7c/aiohttp-3.12.15-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:db71ce547012a5420a39c1b744d485cfb823564d01d5d20805977f5ea1345676", size = 1691634, upload-time = "2025-07-29T05:50:35.846Z" },
-    { url = "https://files.pythonhosted.org/packages/ca/bf/23a335a6670b5f5dfc6d268328e55a22651b440fca341a64fccf1eada0c6/aiohttp-3.12.15-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:ced339d7c9b5030abad5854aa5413a77565e5b6e6248ff927d3e174baf3badf7", size = 1785238, upload-time = "2025-07-29T05:50:37.597Z" },
-    { url = "https://files.pythonhosted.org/packages/57/4f/ed60a591839a9d85d40694aba5cef86dde9ee51ce6cca0bb30d6eb1581e7/aiohttp-3.12.15-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:7c7dd29c7b5bda137464dc9bfc738d7ceea46ff70309859ffde8c022e9b08ba7", size = 1805701, upload-time = "2025-07-29T05:50:39.591Z" },
-    { url = "https://files.pythonhosted.org/packages/85/e0/444747a9455c5de188c0f4a0173ee701e2e325d4b2550e9af84abb20cdba/aiohttp-3.12.15-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:421da6fd326460517873274875c6c5a18ff225b40da2616083c5a34a7570b685", size = 1718758, upload-time = "2025-07-29T05:50:41.292Z" },
-    { url = "https://files.pythonhosted.org/packages/36/ab/1006278d1ffd13a698e5dd4bfa01e5878f6bddefc296c8b62649753ff249/aiohttp-3.12.15-cp311-cp311-win32.whl", hash = "sha256:4420cf9d179ec8dfe4be10e7d0fe47d6d606485512ea2265b0d8c5113372771b", size = 428868, upload-time = "2025-07-29T05:50:43.063Z" },
-    { url = "https://files.pythonhosted.org/packages/10/97/ad2b18700708452400278039272032170246a1bf8ec5d832772372c71f1a/aiohttp-3.12.15-cp311-cp311-win_amd64.whl", hash = "sha256:edd533a07da85baa4b423ee8839e3e91681c7bfa19b04260a469ee94b778bf6d", size = 453273, upload-time = "2025-07-29T05:50:44.613Z" },
-    { url = "https://files.pythonhosted.org/packages/63/97/77cb2450d9b35f517d6cf506256bf4f5bda3f93a66b4ad64ba7fc917899c/aiohttp-3.12.15-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:802d3868f5776e28f7bf69d349c26fc0efadb81676d0afa88ed00d98a26340b7", size = 702333, upload-time = "2025-07-29T05:50:46.507Z" },
-    { url = "https://files.pythonhosted.org/packages/83/6d/0544e6b08b748682c30b9f65640d006e51f90763b41d7c546693bc22900d/aiohttp-3.12.15-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:f2800614cd560287be05e33a679638e586a2d7401f4ddf99e304d98878c29444", size = 476948, upload-time = "2025-07-29T05:50:48.067Z" },
-    { url = "https://files.pythonhosted.org/packages/3a/1d/c8c40e611e5094330284b1aea8a4b02ca0858f8458614fa35754cab42b9c/aiohttp-3.12.15-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:8466151554b593909d30a0a125d638b4e5f3836e5aecde85b66b80ded1cb5b0d", size = 469787, upload-time = "2025-07-29T05:50:49.669Z" },
-    { url = "https://files.pythonhosted.org/packages/38/7d/b76438e70319796bfff717f325d97ce2e9310f752a267bfdf5192ac6082b/aiohttp-3.12.15-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2e5a495cb1be69dae4b08f35a6c4579c539e9b5706f606632102c0f855bcba7c", size = 1716590, upload-time = "2025-07-29T05:50:51.368Z" },
-    { url = "https://files.pythonhosted.org/packages/79/b1/60370d70cdf8b269ee1444b390cbd72ce514f0d1cd1a715821c784d272c9/aiohttp-3.12.15-cp312-cp312-manylinux_2_17_armv7l.manylinux2014_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:6404dfc8cdde35c69aaa489bb3542fb86ef215fc70277c892be8af540e5e21c0", size = 1699241, upload-time = "2025-07-29T05:50:53.628Z" },
-    { url = "https://files.pythonhosted.org/packages/a3/2b/4968a7b8792437ebc12186db31523f541943e99bda8f30335c482bea6879/aiohttp-3.12.15-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:3ead1c00f8521a5c9070fcb88f02967b1d8a0544e6d85c253f6968b785e1a2ab", size = 1754335, upload-time = "2025-07-29T05:50:55.394Z" },
-    { url = "https://files.pythonhosted.org/packages/fb/c1/49524ed553f9a0bec1a11fac09e790f49ff669bcd14164f9fab608831c4d/aiohttp-3.12.15-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:6990ef617f14450bc6b34941dba4f12d5613cbf4e33805932f853fbd1cf18bfb", size = 1800491, upload-time = "2025-07-29T05:50:57.202Z" },
-    { url = "https://files.pythonhosted.org/packages/de/5e/3bf5acea47a96a28c121b167f5ef659cf71208b19e52a88cdfa5c37f1fcc/aiohttp-3.12.15-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fd736ed420f4db2b8148b52b46b88ed038d0354255f9a73196b7bbce3ea97545", size = 1719929, upload-time = "2025-07-29T05:50:59.192Z" },
-    { url = "https://files.pythonhosted.org/packages/39/94/8ae30b806835bcd1cba799ba35347dee6961a11bd507db634516210e91d8/aiohttp-3.12.15-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:3c5092ce14361a73086b90c6efb3948ffa5be2f5b6fbcf52e8d8c8b8848bb97c", size = 1635733, upload-time = "2025-07-29T05:51:01.394Z" },
-    { url = "https://files.pythonhosted.org/packages/7a/46/06cdef71dd03acd9da7f51ab3a9107318aee12ad38d273f654e4f981583a/aiohttp-3.12.15-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:aaa2234bb60c4dbf82893e934d8ee8dea30446f0647e024074237a56a08c01bd", size = 1696790, upload-time = "2025-07-29T05:51:03.657Z" },
-    { url = "https://files.pythonhosted.org/packages/02/90/6b4cfaaf92ed98d0ec4d173e78b99b4b1a7551250be8937d9d67ecb356b4/aiohttp-3.12.15-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:6d86a2fbdd14192e2f234a92d3b494dd4457e683ba07e5905a0b3ee25389ac9f", size = 1718245, upload-time = "2025-07-29T05:51:05.911Z" },
-    { url = "https://files.pythonhosted.org/packages/2e/e6/2593751670fa06f080a846f37f112cbe6f873ba510d070136a6ed46117c6/aiohttp-3.12.15-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:a041e7e2612041a6ddf1c6a33b883be6a421247c7afd47e885969ee4cc58bd8d", size = 1658899, upload-time = "2025-07-29T05:51:07.753Z" },
-    { url = "https://files.pythonhosted.org/packages/8f/28/c15bacbdb8b8eb5bf39b10680d129ea7410b859e379b03190f02fa104ffd/aiohttp-3.12.15-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:5015082477abeafad7203757ae44299a610e89ee82a1503e3d4184e6bafdd519", size = 1738459, upload-time = "2025-07-29T05:51:09.56Z" },
-    { url = "https://files.pythonhosted.org/packages/00/de/c269cbc4faa01fb10f143b1670633a8ddd5b2e1ffd0548f7aa49cb5c70e2/aiohttp-3.12.15-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:56822ff5ddfd1b745534e658faba944012346184fbfe732e0d6134b744516eea", size = 1766434, upload-time = "2025-07-29T05:51:11.423Z" },
-    { url = "https://files.pythonhosted.org/packages/52/b0/4ff3abd81aa7d929b27d2e1403722a65fc87b763e3a97b3a2a494bfc63bc/aiohttp-3.12.15-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:b2acbbfff69019d9014508c4ba0401822e8bae5a5fdc3b6814285b71231b60f3", size = 1726045, upload-time = "2025-07-29T05:51:13.689Z" },
-    { url = "https://files.pythonhosted.org/packages/71/16/949225a6a2dd6efcbd855fbd90cf476052e648fb011aa538e3b15b89a57a/aiohttp-3.12.15-cp312-cp312-win32.whl", hash = "sha256:d849b0901b50f2185874b9a232f38e26b9b3d4810095a7572eacea939132d4e1", size = 423591, upload-time = "2025-07-29T05:51:15.452Z" },
-    { url = "https://files.pythonhosted.org/packages/2b/d8/fa65d2a349fe938b76d309db1a56a75c4fb8cc7b17a398b698488a939903/aiohttp-3.12.15-cp312-cp312-win_amd64.whl", hash = "sha256:b390ef5f62bb508a9d67cb3bba9b8356e23b3996da7062f1a57ce1a79d2b3d34", size = 450266, upload-time = "2025-07-29T05:51:17.239Z" },
-    { url = "https://files.pythonhosted.org/packages/f2/33/918091abcf102e39d15aba2476ad9e7bd35ddb190dcdd43a854000d3da0d/aiohttp-3.12.15-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:9f922ffd05034d439dde1c77a20461cf4a1b0831e6caa26151fe7aa8aaebc315", size = 696741, upload-time = "2025-07-29T05:51:19.021Z" },
-    { url = "https://files.pythonhosted.org/packages/b5/2a/7495a81e39a998e400f3ecdd44a62107254803d1681d9189be5c2e4530cd/aiohttp-3.12.15-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:2ee8a8ac39ce45f3e55663891d4b1d15598c157b4d494a4613e704c8b43112cd", size = 474407, upload-time = "2025-07-29T05:51:21.165Z" },
-    { url = "https://files.pythonhosted.org/packages/49/fc/a9576ab4be2dcbd0f73ee8675d16c707cfc12d5ee80ccf4015ba543480c9/aiohttp-3.12.15-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:3eae49032c29d356b94eee45a3f39fdf4b0814b397638c2f718e96cfadf4c4e4", size = 466703, upload-time = "2025-07-29T05:51:22.948Z" },
-    { url = "https://files.pythonhosted.org/packages/09/2f/d4bcc8448cf536b2b54eed48f19682031ad182faa3a3fee54ebe5b156387/aiohttp-3.12.15-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b97752ff12cc12f46a9b20327104448042fce5c33a624f88c18f66f9368091c7", size = 1705532, upload-time = "2025-07-29T05:51:25.211Z" },
-    { url = "https://files.pythonhosted.org/packages/f1/f3/59406396083f8b489261e3c011aa8aee9df360a96ac8fa5c2e7e1b8f0466/aiohttp-3.12.15-cp313-cp313-manylinux_2_17_armv7l.manylinux2014_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:894261472691d6fe76ebb7fcf2e5870a2ac284c7406ddc95823c8598a1390f0d", size = 1686794, upload-time = "2025-07-29T05:51:27.145Z" },
-    { url = "https://files.pythonhosted.org/packages/dc/71/164d194993a8d114ee5656c3b7ae9c12ceee7040d076bf7b32fb98a8c5c6/aiohttp-3.12.15-cp313-cp313-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:5fa5d9eb82ce98959fc1031c28198b431b4d9396894f385cb63f1e2f3f20ca6b", size = 1738865, upload-time = "2025-07-29T05:51:29.366Z" },
-    { url = "https://files.pythonhosted.org/packages/1c/00/d198461b699188a93ead39cb458554d9f0f69879b95078dce416d3209b54/aiohttp-3.12.15-cp313-cp313-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:f0fa751efb11a541f57db59c1dd821bec09031e01452b2b6217319b3a1f34f3d", size = 1788238, upload-time = "2025-07-29T05:51:31.285Z" },
-    { url = "https://files.pythonhosted.org/packages/85/b8/9e7175e1fa0ac8e56baa83bf3c214823ce250d0028955dfb23f43d5e61fd/aiohttp-3.12.15-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:5346b93e62ab51ee2a9d68e8f73c7cf96ffb73568a23e683f931e52450e4148d", size = 1710566, upload-time = "2025-07-29T05:51:33.219Z" },
-    { url = "https://files.pythonhosted.org/packages/59/e4/16a8eac9df39b48ae102ec030fa9f726d3570732e46ba0c592aeeb507b93/aiohttp-3.12.15-cp313-cp313-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:049ec0360f939cd164ecbfd2873eaa432613d5e77d6b04535e3d1fbae5a9e645", size = 1624270, upload-time = "2025-07-29T05:51:35.195Z" },
-    { url = "https://files.pythonhosted.org/packages/1f/f8/cd84dee7b6ace0740908fd0af170f9fab50c2a41ccbc3806aabcb1050141/aiohttp-3.12.15-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:b52dcf013b57464b6d1e51b627adfd69a8053e84b7103a7cd49c030f9ca44461", size = 1677294, upload-time = "2025-07-29T05:51:37.215Z" },
-    { url = "https://files.pythonhosted.org/packages/ce/42/d0f1f85e50d401eccd12bf85c46ba84f947a84839c8a1c2c5f6e8ab1eb50/aiohttp-3.12.15-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:9b2af240143dd2765e0fb661fd0361a1b469cab235039ea57663cda087250ea9", size = 1708958, upload-time = "2025-07-29T05:51:39.328Z" },
-    { url = "https://files.pythonhosted.org/packages/d5/6b/f6fa6c5790fb602538483aa5a1b86fcbad66244997e5230d88f9412ef24c/aiohttp-3.12.15-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:ac77f709a2cde2cc71257ab2d8c74dd157c67a0558a0d2799d5d571b4c63d44d", size = 1651553, upload-time = "2025-07-29T05:51:41.356Z" },
-    { url = "https://files.pythonhosted.org/packages/04/36/a6d36ad545fa12e61d11d1932eef273928b0495e6a576eb2af04297fdd3c/aiohttp-3.12.15-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:47f6b962246f0a774fbd3b6b7be25d59b06fdb2f164cf2513097998fc6a29693", size = 1727688, upload-time = "2025-07-29T05:51:43.452Z" },
-    { url = "https://files.pythonhosted.org/packages/aa/c8/f195e5e06608a97a4e52c5d41c7927301bf757a8e8bb5bbf8cef6c314961/aiohttp-3.12.15-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:760fb7db442f284996e39cf9915a94492e1896baac44f06ae551974907922b64", size = 1761157, upload-time = "2025-07-29T05:51:45.643Z" },
-    { url = "https://files.pythonhosted.org/packages/05/6a/ea199e61b67f25ba688d3ce93f63b49b0a4e3b3d380f03971b4646412fc6/aiohttp-3.12.15-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:ad702e57dc385cae679c39d318def49aef754455f237499d5b99bea4ef582e51", size = 1710050, upload-time = "2025-07-29T05:51:48.203Z" },
-    { url = "https://files.pythonhosted.org/packages/b4/2e/ffeb7f6256b33635c29dbed29a22a723ff2dd7401fff42ea60cf2060abfb/aiohttp-3.12.15-cp313-cp313-win32.whl", hash = "sha256:f813c3e9032331024de2eb2e32a88d86afb69291fbc37a3a3ae81cc9917fb3d0", size = 422647, upload-time = "2025-07-29T05:51:50.718Z" },
-    { url = "https://files.pythonhosted.org/packages/1b/8e/78ee35774201f38d5e1ba079c9958f7629b1fd079459aea9467441dbfbf5/aiohttp-3.12.15-cp313-cp313-win_amd64.whl", hash = "sha256:1a649001580bdb37c6fdb1bebbd7e3bc688e8ec2b5c6f52edbb664662b17dc84", size = 449067, upload-time = "2025-07-29T05:51:52.549Z" },
+    { url = "https://files.pythonhosted.org/packages/26/dd/bf526e6f0a1120dd6f2df2e97bacfe4d358f13d17a0ff5847301a1375a51/aiohttp-3.14.1-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:aa00140699487bd435fde4342d85c94cb256b7cd3a5b9c3396c67f19922afda2", size = 765225, upload-time = "2026-06-07T21:06:07.957Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/e1/a2872aa55495a70f61310d411541c6ee23812d9a884e000c716e1bc3edbf/aiohttp-3.14.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:1c1af67559445498b502030c35c59db59966f47041ca9de5b4e707f86bd10b5f", size = 518743, upload-time = "2026-06-07T21:06:09.749Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/e7/c60c7b209e509cc787de3cea0550a518538cfc08003e1c1e14c1c63fff71/aiohttp-3.14.1-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:d44ec478e713ee7f29b439f7eb8dc2b9d4079e11ae114d2c2ac3d5daf30516c8", size = 514139, upload-time = "2026-06-07T21:06:11.26Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/8d/614ace2f579702c9840ab1e1447fd8509e35b0b904f7196418fa2f57b25d/aiohttp-3.14.1-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d3b1a184a9a8f548a6b73f1e26b96b052193e4b3175ed7342aaf1151a1f00a04", size = 1784088, upload-time = "2026-06-07T21:06:12.887Z" },
+    { url = "https://files.pythonhosted.org/packages/49/e0/726e90f99542bf292f81a96a12cc4847deb86f3ccf62c6f4014a201f4d33/aiohttp-3.14.1-cp311-cp311-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:5f2504bc0322437c9a1ff6d3333ca56c7477b727c995f036b976ae17b98372c8", size = 1737835, upload-time = "2026-06-07T21:06:14.564Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/4b/d176d5c4db9d33dacf0543102ea59503bc1d528af4cfd0b719949ca49389/aiohttp-3.14.1-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:73f05ea02013e02512c3bf42714f1208c57168c779cc6fe23516e4543089d0a6", size = 1842801, upload-time = "2026-06-07T21:06:16.228Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/d6/5a99b563690ea0cbed912ae94a2ce33993a5709a651a3a4fe761e7dd973a/aiohttp-3.14.1-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:797457503c2d426bee06eef808d07b31ede30b65e054444e7de64cad0061b7af", size = 1929992, upload-time = "2026-06-07T21:06:17.947Z" },
+    { url = "https://files.pythonhosted.org/packages/76/7f/a987b14a3859094b3cea3f4825219c3e5536242564af6e3f9c2f6c994eb2/aiohttp-3.14.1-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:b821a1f7dedf7e37450654e620038ac3b2e81e8fa6ea269337e97101978ec730", size = 1786989, upload-time = "2026-06-07T21:06:19.677Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/1a/420e5c85a3e73349372ed22ce0b6af86bfa6ce16a4b20a64a2e94608c781/aiohttp-3.14.1-cp311-cp311-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:4cd96b5ba05d67ed0cf00b5b405c8cd99586d8e3481e8ee0a831057591af7621", size = 1640129, upload-time = "2026-06-07T21:06:22.558Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/80/18a592ed3be0a402cc03670bd72ee1f8563ddbe1d8d5542dbf868f274136/aiohttp-3.14.1-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:1d459b98a932296c6f0e94f87511a0b1b90a8a02c30a50e60a297619cd5a58ee", size = 1756576, upload-time = "2026-06-07T21:06:24.8Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/0b/8b3d5713373858ff71a617daf6e3b0e81ad63e79d09a3cf2f6b6b983939c/aiohttp-3.14.1-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:764457a7be60825fb770a644852ff717bcbb5042f189f2bd16df61a81b3f6573", size = 1754668, upload-time = "2026-06-07T21:06:26.528Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/49/fd564575cf225821d7ba5a117cb8bc27213d8a7e1811162afb43ae077039/aiohttp-3.14.1-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:f7a16ef45b081454ef844502d87a848876c490c4cb5c650c230f6ec79ed2c1e7", size = 1817019, upload-time = "2026-06-07T21:06:28.297Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/1b/e850c9ae6fc91356552ae668bb6c51e93fa29c8aef13398a10b56678557f/aiohttp-3.14.1-cp311-cp311-musllinux_1_2_riscv64.whl", hash = "sha256:2fbc3ed048b3475b9f0cbcb9978e9d2d3511acd91ead203af26ed9f0056004cf", size = 1631638, upload-time = "2026-06-07T21:06:30.242Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/94/3c337ba72451a89806ace6f75bddc92bafc5b8d53d90115a512858024b63/aiohttp-3.14.1-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:bedb0cd073cc2dc035e30aeb99444389d3cd2113afe4ef9fcd23d439f5bade85", size = 1835660, upload-time = "2026-06-07T21:06:31.943Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/9c/9c18cf367a0498212d9ba7daf990b504a5e8ae064cda4b504e2647c89c03/aiohttp-3.14.1-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:b6feea921016eb3d4e04d65fc4e9ca402d1a3801f562aef94989f54694917af3", size = 1775698, upload-time = "2026-06-07T21:06:33.72Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/63/a251a9d2a6cb45065b2ddc0bde2b3dd10108740a9a42f632c66405a761a2/aiohttp-3.14.1-cp311-cp311-win32.whl", hash = "sha256:313701e488100074ce99850404ee36e741abf6330179fec908a1944ecf570126", size = 458386, upload-time = "2026-06-07T21:06:35.279Z" },
+    { url = "https://files.pythonhosted.org/packages/17/ca/69274c51dcd6e8947d77b2806cf47a4a15f2c846e2cbeb1882547d3da283/aiohttp-3.14.1-cp311-cp311-win_amd64.whl", hash = "sha256:03ab4530fdcb3a543a122ba4b65ac9919da9fe9f78a03d328a6e38ff962f7aa5", size = 483406, upload-time = "2026-06-07T21:06:36.824Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/8a/c25904f77690c3688ec140f87591ef11a0cfe36bf3d5c0f1f38056fb62b3/aiohttp-3.14.1-cp311-cp311-win_arm64.whl", hash = "sha256:486f7d16ed54c39c2cbd7ca71fd8ba2b8bb7860df65bd7b6ed640bab96a38a8b", size = 452987, upload-time = "2026-06-07T21:06:38.371Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/21/151624b51cd92553d95424daf4bf19f19ce9be9002d19253e7e7ce67197b/aiohttp-3.14.1-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:d35143e27778b4bb0fb189562d7f275bff79c62ab8e98459717c0ea617ff2480", size = 757402, upload-time = "2026-06-07T21:06:40.311Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/82/280619e0bd7bf2454987e19282616e84762255dd9c8468f62382e8c191f1/aiohttp-3.14.1-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:bcfb80a2cc36fba2534e5e5b5264dc7ae6fcd9bf15256da3e53d2f499e6fa29d", size = 512310, upload-time = "2026-06-07T21:06:42.207Z" },
+    { url = "https://files.pythonhosted.org/packages/55/b2/2aac325583aaa1353045f96dffa586d8a34e8322e14a7ba49cffeb103ab4/aiohttp-3.14.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:27fd7c91e51729b4f7e1577865fa6d34c9adccbc39aabe9000285b48af9f0ec2", size = 512448, upload-time = "2026-06-07T21:06:43.813Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/72/a60607cb849faa8af8a356c9329ea2eb6f395d49e82cc82ccba1fd8deb8f/aiohttp-3.14.1-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:64c567bf9eaf664280116a8688f63016e6b32db2505908e2bdaca1b6438142f2", size = 1766854, upload-time = "2026-06-07T21:06:45.391Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/d3/d9fe1c9ec7557ab4d0d82bebaa728c6418f0b93295ec2f4ab015f7710cc7/aiohttp-3.14.1-cp312-cp312-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:f5e6ff2bdbb8f4cd3fbe41f99e25bbcd58e3bf9f13d3dd31a11e7917251cc77a", size = 1740884, upload-time = "2026-06-07T21:06:47.413Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/dc/f2cecfaf9337ba3e63f181500814ff502aa3d00d9c7ec93a9d23d10a27b2/aiohttp-3.14.1-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:2f73e01dc37122325caf079982621262f96d74823c179038a82fddfc50359264", size = 1810034, upload-time = "2026-06-07T21:06:50.165Z" },
+    { url = "https://files.pythonhosted.org/packages/66/d7/2ff65c5e65c0d7476daf7e15c032e0805e36811185b9623e3238ad6c763e/aiohttp-3.14.1-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:bb2c0c80d431c0d03f2c7dbf125150fedd4f0de17366a7ca33f7ccb822391842", size = 1904054, upload-time = "2026-06-07T21:06:52.035Z" },
+    { url = "https://files.pythonhosted.org/packages/20/9c/d445818389df371f56d141d881153ba23183c4735a03f7356ffb43f7757d/aiohttp-3.14.1-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:3e6fc1a85fa7194a1a7d19f44e8609180f4a8eb5fa4c7ed8b4355f080fad235c", size = 1790278, upload-time = "2026-06-07T21:06:54.049Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/aa/bf04cb4d865fc6101c2229a294ad744973b72e513fdc5a6b791e6983d72a/aiohttp-3.14.1-cp312-cp312-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:686b6c0d3911ec387b444ddf5dc62fb7f7c0a7d5186a7861626496a5ab4aff95", size = 1591795, upload-time = "2026-06-07T21:06:55.911Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/b4/4dac0038960427ba832f6609dfb4ea5437d7fd80c72001b9e48f834f428b/aiohttp-3.14.1-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:c6fa4dc7ad6f8109c70bb1499e589f76b0b792baf39f9b017eb92c8a81d0a199", size = 1728397, upload-time = "2026-06-07T21:06:57.777Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/f9/7cd4e8ad7aa3b75f17d56bb5498dd604a93d4e6eece822ba0568c413fff0/aiohttp-3.14.1-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:87a5eea1b2a5e21e1ebdbb33ad4165359189327e63fc4e4894693e7f821ac817", size = 1766504, upload-time = "2026-06-07T21:07:00.009Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/df/fc01d9fcad0f73fed3f3d361f1f94f975947b50dff82919f6dc2bf4316cc/aiohttp-3.14.1-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:1c1421eb01d4fd608d88cc8290211d177a58532b55ad94076fb349c5bf467f0a", size = 1777806, upload-time = "2026-06-07T21:07:02.064Z" },
+    { url = "https://files.pythonhosted.org/packages/41/09/47e2d090bddcc8fb4ccb4c314aadc32d7c5d9bb55f50f6ad1c92fc15d501/aiohttp-3.14.1-cp312-cp312-musllinux_1_2_riscv64.whl", hash = "sha256:34b257ec41345c1e8f2df68fa908a7952f5de932723871eb633ecbbff396c9a4", size = 1580707, upload-time = "2026-06-07T21:07:03.942Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/36/f1a4ce904ae0b6930cfe9afc96d0896f7ec1a620c400405d63783bb95a9c/aiohttp-3.14.1-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:de538791a80e5d862addbc183f70f0158ac9b9bb872bb147f1fd2a683691e087", size = 1798121, upload-time = "2026-06-07T21:07:05.987Z" },
+    { url = "https://files.pythonhosted.org/packages/70/0a/e0075ce9ca0279ee1d4f0c0b85f54fea02ebc83c3007651a72bece658fec/aiohttp-3.14.1-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:6f71173be42d3241d428f760122febb748de0623f44308a6f120d0dd9ec572e3", size = 1767580, upload-time = "2026-06-07T21:07:07.873Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/61/a0c0a8f327a9c52095cdd8e312391b00d3ed64ab6c72bb5c33d8ec251cf7/aiohttp-3.14.1-cp312-cp312-win32.whl", hash = "sha256:ec8dc383ee57ea3e883477dcca3f11b65d58199f1080acaf4cd6ad9a99698be4", size = 452771, upload-time = "2026-06-07T21:07:09.669Z" },
+    { url = "https://files.pythonhosted.org/packages/df/d9/ea367c75f16ac9c6cdc8febb25e8318fa21a2b1bc8d6514d4b2d890bface/aiohttp-3.14.1-cp312-cp312-win_amd64.whl", hash = "sha256:2aa92c87868cd13674989f9ee83e5f9f7ea4237589b728048e1f0c8f6caa3271", size = 479873, upload-time = "2026-06-07T21:07:11.538Z" },
+    { url = "https://files.pythonhosted.org/packages/03/64/8d96784a7851156db8a4c6c3f6f91042fdf39fb15a4cc38c8b3c14833c45/aiohttp-3.14.1-cp312-cp312-win_arm64.whl", hash = "sha256:2c840c90759922cb5e6dda94596e079a30fb5a5ba548e7e0dc00574703940847", size = 448073, upload-time = "2026-06-07T21:07:13.637Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/97/bd137012dd97e1649162b099135a80e1fd59aaa807b2430fc448d1029aff/aiohttp-3.14.1-cp313-cp313-android_21_arm64_v8a.whl", hash = "sha256:b3a03285a7f9c7b016324574a6d92a1c895da6b978cb8f1deee3ac72bc6da178", size = 506882, upload-time = "2026-06-07T21:07:15.501Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/79/e5cc690e9d922a66887ceeaca53a8ffd5a7b0be3816142b7abc433742d89/aiohttp-3.14.1-cp313-cp313-android_21_x86_64.whl", hash = "sha256:2a73f487ab8ef5abbb24b7aa9b73e98eaba9e9e031804ff2416f02eca315ccaf", size = 515270, upload-time = "2026-06-07T21:07:17.53Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/22/a73ccbf9dbd6e26dda0b24d5fd5db7da92ee3383a79f47677ffb834c5c5b/aiohttp-3.14.1-cp313-cp313-ios_13_0_arm64_iphoneos.whl", hash = "sha256:915fbb7b41b115192259f8c9ae58f3ddc444d2b5579917270211858e606a4afd", size = 485841, upload-time = "2026-06-07T21:07:19.555Z" },
+    { url = "https://files.pythonhosted.org/packages/3b/b9/57ed8eaf596321c2ad747bd480fb1700dbd7177c60dfc9e4c187f629662e/aiohttp-3.14.1-cp313-cp313-ios_13_0_arm64_iphonesimulator.whl", hash = "sha256:7fb4bdf95b0561a79f259f9d28fbc109728c5ee7f27aff6391f0ca703a329abe", size = 492088, upload-time = "2026-06-07T21:07:21.581Z" },
+    { url = "https://files.pythonhosted.org/packages/78/c0/5ebe5270a7c140d7c6f79dcb018640225f14d406c149e4eec04a7d82fe71/aiohttp-3.14.1-cp313-cp313-ios_13_0_x86_64_iphonesimulator.whl", hash = "sha256:1b9748363260121d2927704f5d4fc498150669ca3ae93625986ee89c8f80dcd4", size = 501564, upload-time = "2026-06-07T21:07:23.388Z" },
+    { url = "https://files.pythonhosted.org/packages/75/7f/8cdaa24fc7983865e0915153b96a9ac5bcdd3548d64c5a27d17cecccad2d/aiohttp-3.14.1-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:86a6dab78b0e43e2897a3bbe15745aa60dc5423ca437b7b0b164c069bf91b876", size = 751998, upload-time = "2026-06-07T21:07:25.046Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/f4/c4227aacfacc5cb0cc2d119b65301d177912a6842cd64e120c47af76064f/aiohttp-3.14.1-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:4dfd6e47d3c44c2279907607f73a4240b88c69eb8b90da7e2441a8045dfd21da", size = 510918, upload-time = "2026-06-07T21:07:27.28Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/01/a2d5f96cd4e74424864d30bc0a7e44d0a12dacdcfa91b5b2d1bd3dca6bf3/aiohttp-3.14.1-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:317acd9f8602858dc7d59679812c376c7f0b97bcbbf16e0d6237f54141d8a8a6", size = 508657, upload-time = "2026-06-07T21:07:29.252Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/ed/3c0fb5c500fdd8e7ebc10d1889c04384fffa1a9163eac1356088ca9da1b1/aiohttp-3.14.1-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:bd869c427324e5cb15195793de951295710db28be7d818247f3097b4ab5d4b96", size = 1757907, upload-time = "2026-06-07T21:07:31.03Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/ab/d4c924d9bd5be3050c226612413ce68cb54c70d2c31b661bfc8d9a5b6a70/aiohttp-3.14.1-cp313-cp313-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:93b032b5ec3255473c143627d21a69ac74ae12f7f33974cb587c564d11b1066f", size = 1737565, upload-time = "2026-06-07T21:07:33.031Z" },
+    { url = "https://files.pythonhosted.org/packages/19/2a/37326821ff779084020cdc33224d20b19f42f4183a500ff92022a739eda7/aiohttp-3.14.1-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:f234b4deb12f3ad59127e037bc57c40c21e45b45282df7d3a55a0f409f595296", size = 1799018, upload-time = "2026-06-07T21:07:35.003Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/4f/6e947ba73e4ce09070761c05ed3a8ceb7c21f5e46798671d8b2aac0e4626/aiohttp-3.14.1-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:9af6779bfb46abf124068327abcdf9ce95c9ef8287a3e8da76ccf2d0f16c28fa", size = 1894416, upload-time = "2026-06-07T21:07:36.956Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/6e/dbf1d0625dc711fb2851f4f3c3055c39ed58bae92082d8c627dbe6013736/aiohttp-3.14.1-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:faccab372e66bc76d5731525e7f1143c922271725b9d38c9f97edcc66266b451", size = 1783881, upload-time = "2026-06-07T21:07:39.063Z" },
+    { url = "https://files.pythonhosted.org/packages/44/c2/5e25098a67268ed369483ae7d1a58bd0a13d03aab860d2a0e4a6eb25b046/aiohttp-3.14.1-cp313-cp313-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:f380468b09d2a81633ee863b0ec5648d364bd17bb8ecfb8c2f387f7ac1faf42c", size = 1587572, upload-time = "2026-06-07T21:07:41.058Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/bd/cf9cee17e140f942a3de73e658a543aa8fbf35a5fc67a9d2538d52d77f0b/aiohttp-3.14.1-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:97e704dcd26271f5bda3fa07c3ce0fb76d6d3f8659f4baa1a24442cc9ba177ca", size = 1722137, upload-time = "2026-06-07T21:07:43.014Z" },
+    { url = "https://files.pythonhosted.org/packages/89/6d/5684f8c59045c96f81a18cefbc1fbbd79d25b88f1c622f2a5c5c08fcb632/aiohttp-3.14.1-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:269b76ac5394092b95bc4a098f4fc6c191c083c3bd12775d1e30e663132f6a09", size = 1755953, upload-time = "2026-06-07T21:07:45.933Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/40/35caf3170f8359760740a7d9aa0fff2e344bef98e1d1186f5a0f6dec17e6/aiohttp-3.14.1-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:5c0b3e614340c889d575451696374c9d17affd54cd607ca0babed8f8c37b9397", size = 1766479, upload-time = "2026-06-07T21:07:48.047Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/a1/b0c61e7a137f0d81de49a82023a6df73c3c16d6fefb0f8e4a93d21639002/aiohttp-3.14.1-cp313-cp313-musllinux_1_2_riscv64.whl", hash = "sha256:5663ee9257cfa1add7253a7da3035a02f31b6600ec48261585e1800a81533080", size = 1580077, upload-time = "2026-06-07T21:07:50.069Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/41/194ea4623693009fcefebef7aef63c141754f153e9cd0d39d3b9e36c175c/aiohttp-3.14.1-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:603a2c834142172ffddc054067f5ec0ca65d57a0aa98a71bc81952573208e345", size = 1791688, upload-time = "2026-06-07T21:07:52.106Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/45/4de841f005cfe1fd63e2a2fe011262c515e2a62aa6994b15947e7d717ac9/aiohttp-3.14.1-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:cb21957bb8aca671c1765e32f58164cf0c50e6bf41c0bbbd16da20732ecaf588", size = 1761094, upload-time = "2026-06-07T21:07:54.113Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/ae/dbce10533d3896d544d5053939ed75b7dc31a1b0973d959b1b5ae21028d6/aiohttp-3.14.1-cp313-cp313-win32.whl", hash = "sha256:e509a55f681e6158c20f70f102f9cf61fb20fbc382272bc6d94b7343f2582780", size = 452662, upload-time = "2026-06-07T21:07:56.06Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/d9/0bf1a19362c32f06229da5e7ddfcec91f93474d6307f7a2d3135e9c674dc/aiohttp-3.14.1-cp313-cp313-win_amd64.whl", hash = "sha256:1ac8531b638959718e18c2207fbfe297819875da46a740b29dfa29beba64355a", size = 479748, upload-time = "2026-06-07T21:07:58.319Z" },
+    { url = "https://files.pythonhosted.org/packages/22/0a/62e7232dc9484fbec112ceb32efb6a624cc7994ec6e2b019286f17c4e8f2/aiohttp-3.14.1-cp313-cp313-win_arm64.whl", hash = "sha256:250d14af67f6b6a1a4a811049b1afa69d61d617fca6bf33149b3ab1a6dbcf7b8", size = 447723, upload-time = "2026-06-07T21:08:00.154Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/a1/5fafa04e1ca91ddb47608699d60649c1c6db3cf41c99e78fc4056f9513db/aiohttp-3.14.1-cp314-cp314-android_24_arm64_v8a.whl", hash = "sha256:7c106c26852ca1c2047c6b80384f17100b4e439af276f21ef3d4e2f450ae7e15", size = 508531, upload-time = "2026-06-07T21:08:02.093Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/2e/bfa02f699d87ffc86d5959270b28f1cb410add3ccaced8ed2e0b8a5238fc/aiohttp-3.14.1-cp314-cp314-android_24_x86_64.whl", hash = "sha256:20205f7f5ade7aaec9f4b500549bbc071b046453aed72f9c06dcab87896a83e8", size = 514718, upload-time = "2026-06-07T21:08:04.476Z" },
+    { url = "https://files.pythonhosted.org/packages/85/a5/9594ad6289eebbc97d167c44213d557807f90e59115caad24de21ad2c3b1/aiohttp-3.14.1-cp314-cp314-ios_13_0_arm64_iphoneos.whl", hash = "sha256:62a759436b29e677181a9e76bab8b8f689a29cb9c535f45f7c48c9c830d3f8c3", size = 487918, upload-time = "2026-06-07T21:08:06.377Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/61/16a32c36c3c49edec122a3dc811f2057df2f94d3b14aa107c8017d981618/aiohttp-3.14.1-cp314-cp314-ios_13_0_arm64_iphonesimulator.whl", hash = "sha256:2964cbf553df4d7a57348da44d961d871895fc1ee4e8c322b2a95612c7b17fba", size = 494014, upload-time = "2026-06-07T21:08:08.263Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/89/3ebcf96ed99c05bec9c434aaac6963fd3cbab4a786ae739908a144d9ce44/aiohttp-3.14.1-cp314-cp314-ios_13_0_x86_64_iphonesimulator.whl", hash = "sha256:237651caadc3a59badd39319c54642b5299e9cc98a3a194310e55d5bb9f5e397", size = 502398, upload-time = "2026-06-07T21:08:10.244Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/3d/b74870a0c2d40c355928cd5b96c7a11fa821b8a40fc41365e64479b151fb/aiohttp-3.14.1-cp314-cp314-macosx_10_15_universal2.whl", hash = "sha256:896e12dfdbbab9d8f7e16d2b28c6769a60126fa92095d1ebf9473d02593a2448", size = 758018, upload-time = "2026-06-07T21:08:12.447Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/66/f42f5c984d99e49c6cff5f26f590750f2e2f7ef1fcfb99966ab5be1b632e/aiohttp-3.14.1-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:d03f281ed22579314ba00821ce20115a7c0ac430660b4cc05704a3f818b3e004", size = 512462, upload-time = "2026-06-07T21:08:14.624Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/a7/248e1aebe0c7810b0271e021a0f2a5eb6e78a051885b3c9df49f42a5802d/aiohttp-3.14.1-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:07eabb979d236335fed927e137a928c9adfb7df3b9ec7aa31726f133a62be983", size = 512824, upload-time = "2026-06-07T21:08:16.572Z" },
+    { url = "https://files.pythonhosted.org/packages/26/97/2aa0e5ba0727dc3bd5aaebb7ccbc510f7dfb7fb961ec87497cd496635ab1/aiohttp-3.14.1-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4fe1f1087cbadb280b5e1bb054a4f00d1423c74d6626c5e48400d871d34ecefe", size = 1749898, upload-time = "2026-06-07T21:08:18.635Z" },
+    { url = "https://files.pythonhosted.org/packages/00/8d/e97f6c96c891d457c8479d92a514ba194d0412f981d72c70341ee18488ed/aiohttp-3.14.1-cp314-cp314-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:367a9314fdc79dab0fac96e216cb41dd73c85bdca85306ce8999118ba7e0f333", size = 1710114, upload-time = "2026-06-07T21:08:20.892Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/e6/aa8d7e863048c8fceb5cd6ce74017311cec3ead07847387e12265fb4444e/aiohttp-3.14.1-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:a24f677ebe83749039e7bdf862ff0bbb16818ae4193d4ef96505e269375bcce0", size = 1802541, upload-time = "2026-06-07T21:08:23.044Z" },
+    { url = "https://files.pythonhosted.org/packages/83/a8/72193137de57fda4ebfae4563182d082c8856e3b6e9871d0b46f028fb369/aiohttp-3.14.1-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:c83afe0ba876be7e943d2e0ba645809ad441575d2840c895c21ee5de93b9377a", size = 1875776, upload-time = "2026-06-07T21:08:25.288Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/18/938441025db6769a3464596b2410af3afde0b21eb2f204c6f766f68af4bd/aiohttp-3.14.1-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:634e385930fb6d2d479cf3aa66515955863b77a5e3c2b5894ca259a25b308602", size = 1760329, upload-time = "2026-06-07T21:08:27.363Z" },
+    { url = "https://files.pythonhosted.org/packages/60/29/bf2496b4065e76e09fe48015aaffe5ce161d8f089b06ac6982070f653076/aiohttp-3.14.1-cp314-cp314-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:eeea07c4397bbc57719c4eed8f9c284874d4f175f9b6d57f7a1546b976d455ca", size = 1587293, upload-time = "2026-06-07T21:08:29.805Z" },
+    { url = "https://files.pythonhosted.org/packages/49/a2/2136674d52123b1354bd05dd5753c318db47dc0c927cc70b27bab3755456/aiohttp-3.14.1-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:335c0cc3e3545ce98dcb9cfcb836f40c3411f43fa03dab757597d80c89af8a35", size = 1714756, upload-time = "2026-06-07T21:08:32.094Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/b9/e5fd2e6f915503081c0f9b1e8540947037929c70c191da2e4d54b31a21a1/aiohttp-3.14.1-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:ae6be797afdef264e8a84864a85b196ca06045586481b3df8a967322fd2fa844", size = 1721052, upload-time = "2026-06-07T21:08:34.167Z" },
+    { url = "https://files.pythonhosted.org/packages/63/5a/2833e324a2263e104e31e2e91bc5bbee81bc499afd32203faee048a883f0/aiohttp-3.14.1-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:8560b4d712474335d08907db7973f71912d3a9a8f1dee992ec06b5d2fe359496", size = 1766888, upload-time = "2026-06-07T21:08:36.95Z" },
+    { url = "https://files.pythonhosted.org/packages/57/fa/dea6511870913162f3b2e8c42a7614eb203a4540b8c2da43e0bfb0548f3c/aiohttp-3.14.1-cp314-cp314-musllinux_1_2_riscv64.whl", hash = "sha256:2b7edd08e0a5deb1e8564a2fcd8f4561014a3f05252334671bbf55ddd47db0e5", size = 1581679, upload-time = "2026-06-07T21:08:39.292Z" },
+    { url = "https://files.pythonhosted.org/packages/14/bd/3cf0d55e71784b33534e9710a67d382d900598b4787fbce6cc7317f8c42a/aiohttp-3.14.1-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:b6ff7fcee63287ae57b5df3e4f5957ce032122802509246dec1a5bcc55904c95", size = 1782021, upload-time = "2026-06-07T21:08:41.407Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/af/14bb5843eccbe234f4dfb78ab73e549d99727247e62ae5d62cbd22eaf5b0/aiohttp-3.14.1-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:6ffbb2f4ec1ceaff7e07d43922954da26b223d188bf30658e561b98e23089444", size = 1742574, upload-time = "2026-06-07T21:08:43.795Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/1e/fbeb7af9210a67ac0f9c9bec0f8f4568497924e33137a3d5b48e1cf85f3f/aiohttp-3.14.1-cp314-cp314-win32.whl", hash = "sha256:a9875b46d910cff3ea2f5962f9d266b465459fe634e22556ab9bd6fc1192eea0", size = 457773, upload-time = "2026-06-07T21:08:46.168Z" },
+    { url = "https://files.pythonhosted.org/packages/f0/2b/13e8d741a9ec5db7d900c060554cf8352ab85e44e2a4469ebb9d377bda17/aiohttp-3.14.1-cp314-cp314-win_amd64.whl", hash = "sha256:af8b4b81a960eeaf1234971ac3cd0ba5901f3cd42eae42a46b4d089a8b492719", size = 485001, upload-time = "2026-06-07T21:08:48.401Z" },
+    { url = "https://files.pythonhosted.org/packages/df/30/491acfa2c4d6c3ff59c49a14fc1b50be3241e25bbb0c84c09e2da4d11395/aiohttp-3.14.1-cp314-cp314-win_arm64.whl", hash = "sha256:cf4491381b1b57425c315a56a439251b1bdac07b2275f19a8c44bc57744532ec", size = 453809, upload-time = "2026-06-07T21:08:50.7Z" },
+    { url = "https://files.pythonhosted.org/packages/34/e3/19dbe1a1f4cc6230eb9e314de7fe68053b0992f9302b27d12141a0b5db53/aiohttp-3.14.1-cp314-cp314t-macosx_10_15_universal2.whl", hash = "sha256:819c054312f1af92947e6a55883d1b66feefab11531a7fc45e0fb9b63880b5c2", size = 793320, upload-time = "2026-06-07T21:08:52.775Z" },
+    { url = "https://files.pythonhosted.org/packages/7f/20/1b7182219ba1b108430d6e4dc53d25ae02dcfcf5a045b33af4e8c5167527/aiohttp-3.14.1-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:10ee9c1753a8f706345b22496c79fbddb5be0599e0823f3738b1534058e25340", size = 529077, upload-time = "2026-06-07T21:08:55Z" },
+    { url = "https://files.pythonhosted.org/packages/b9/c8/14ce60ec31a2e5f5274bb17d383a6f7a3aabca31ac04eee05585bbadab16/aiohttp-3.14.1-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:1601cc37baf5750ccacae618ec2daf020769581695550e3b654a911f859c563d", size = 532476, upload-time = "2026-06-07T21:08:57.176Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/02/9ac85e081e53da2e061b02fa7758fe0a12d17b8ce2d1f5e6c7cb76730328/aiohttp-3.14.1-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4d6e0ac9da31c9c04c84e1c0182ad8d6df35965a85cae29cd71d089621b3ae94", size = 1922347, upload-time = "2026-06-07T21:08:59.563Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/3e/d3ba07a0ab38b5389e10bec4362d21e10a4f667cba2d79ba30837b3a5059/aiohttp-3.14.1-cp314-cp314t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:9e8f2d660c350b3d0e259c7a7e3d9b7fc8b41210cbcc3d4a7076ff0a5e5c2fdc", size = 1786465, upload-time = "2026-06-07T21:09:01.909Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/cb/e2ee978a00cfb2df829704a69528b18154eba5939f45bc1efa8f33aee4c5/aiohttp-3.14.1-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:4691802dda97be727f79d86818acaad7eb8e9252626a1d6b519fedbb92d5e251", size = 1909423, upload-time = "2026-06-07T21:09:04.357Z" },
+    { url = "https://files.pythonhosted.org/packages/73/5d/1430334858b1022b58ae50399a918f0bd6fe8fa7fa183598d657ff61e040/aiohttp-3.14.1-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:c389c482a7e9b9dc3ee2701ac46c4125297a3818875b9c305ddb603c04828fd1", size = 2001906, upload-time = "2026-06-07T21:09:06.722Z" },
+    { url = "https://files.pythonhosted.org/packages/66/4e/560c7472d3d198a23aa5c8b19a5115bf6a9b77b7d3e4bb363da320430ad2/aiohttp-3.14.1-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:fc0cacab7ba4e56f0f81c82a98c09bed2f39c940107b03a34b168bdf7597edd3", size = 1877095, upload-time = "2026-06-07T21:09:09.011Z" },
+    { url = "https://files.pythonhosted.org/packages/0d/f1/4745806578d447db4a784a8591e2dae3afdfc2bcb96f8f81271b13df6543/aiohttp-3.14.1-cp314-cp314t-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:979ed4717f59b8bb12e3963378fa285d93d367e15bcd66c721311826d3c44a6c", size = 1676222, upload-time = "2026-06-07T21:09:11.461Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/c9/48255813cca749a229ef0ab476004ec623728ad79a9c0840616f6c076325/aiohttp-3.14.1-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:38e1e7daaea81df51c952e18483f323d878499a1e2bfe564790e0f9701d6f203", size = 1842922, upload-time = "2026-06-07T21:09:14.118Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/c0/bbd054e2bee909f529523a5af3891052606af5143c09f5f183ec3b234676/aiohttp-3.14.1-cp314-cp314t-musllinux_1_2_armv7l.whl", hash = "sha256:4132e72c608fe9fecb8f409113567605915b83e9bdd3ea56538d2f9cd35002f1", size = 1825035, upload-time = "2026-06-07T21:09:16.447Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/ae/90395d4376deceb74e09ec26b6adf7d2015a6f8802d6d84446af860fef04/aiohttp-3.14.1-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:eefd9cc9b6d4a2db5f00a26bc3e4f9acf71926a6ec557cd56c9c6f27c290b665", size = 1849512, upload-time = "2026-06-07T21:09:18.742Z" },
+    { url = "https://files.pythonhosted.org/packages/93/bd/fb25f3049957553d4ce0ba6ae480aa2f592a6985497fca590837d16c1be0/aiohttp-3.14.1-cp314-cp314t-musllinux_1_2_riscv64.whl", hash = "sha256:b165790117eea512d7f3fb22f1f6dad3d55a7189571993eb015591c1401276d1", size = 1668571, upload-time = "2026-06-07T21:09:21.458Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/22/7f73303d64dd567ff3addca90b556690ed1233a47b8f55d242fb90af3681/aiohttp-3.14.1-cp314-cp314t-musllinux_1_2_s390x.whl", hash = "sha256:ed09c7eb1c391271c2ed0314a51903e72a3acb653d5ccfc264cdf3ef11f8269d", size = 1881159, upload-time = "2026-06-07T21:09:23.813Z" },
+    { url = "https://files.pythonhosted.org/packages/44/be/0474c5a8b5640e1e4aa1923430a91f4151be82e511373fe764189b89aef5/aiohttp-3.14.1-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:99abd37084b82f5830c635fddd0b4993b9742a66eb746dacf433c8590e8f9e3c", size = 1841409, upload-time = "2026-06-07T21:09:26.207Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/3c/bb4a7cba26956cb3da4553cc2056cf67be5b5ff6e6d8fa4fbdff73bfb7ae/aiohttp-3.14.1-cp314-cp314t-win32.whl", hash = "sha256:47ddf841cdecc810749921d25606dee45857d12d2ad5ddb7b5bd7eab12e4b365", size = 494166, upload-time = "2026-06-07T21:09:28.505Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/84/ec80c2c1f66a952555a9f86df6b33af65108a6febfa0471b69013a12f807/aiohttp-3.14.1-cp314-cp314t-win_amd64.whl", hash = "sha256:5e78b522b7a6e27e0b25d19b247b75039ac4c94f99823e3c9e53ae1603a9f7e9", size = 530255, upload-time = "2026-06-07T21:09:30.843Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/71/6e22be134a4061ada85a92951b842f2657f17d926b727f3f94c56ae963d6/aiohttp-3.14.1-cp314-cp314t-win_arm64.whl", hash = "sha256:90d53f1609c29ccc2193945ef732428382a28f78d0456ae4d3daf0d48b74f0f6", size = 469640, upload-time = "2026-06-07T21:09:33.028Z" },
 ]
 
 [[package]]
@@ -261,56 +311,50 @@ wheels = [
 
 [[package]]
 name = "brotli"
-version = "1.1.0"
+version = "1.2.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/2f/c2/f9e977608bdf958650638c3f1e28f85a1b075f075ebbe77db8555463787b/Brotli-1.1.0.tar.gz", hash = "sha256:81de08ac11bcb85841e440c13611c00b67d3bf82698314928d0b676362546724", size = 7372270, upload-time = "2023-09-07T14:05:41.643Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/f7/16/c92ca344d646e71a43b8bb353f0a6490d7f6e06210f8554c8f874e454285/brotli-1.2.0.tar.gz", hash = "sha256:e310f77e41941c13340a95976fe66a8a95b01e783d430eeaf7a2f87e0a57dd0a", size = 7388632, upload-time = "2025-11-05T18:39:42.86Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/96/12/ad41e7fadd5db55459c4c401842b47f7fee51068f86dd2894dd0dcfc2d2a/Brotli-1.1.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:a3daabb76a78f829cafc365531c972016e4aa8d5b4bf60660ad8ecee19df7ccc", size = 873068, upload-time = "2023-09-07T14:03:37.779Z" },
-    { url = "https://files.pythonhosted.org/packages/95/4e/5afab7b2b4b61a84e9c75b17814198ce515343a44e2ed4488fac314cd0a9/Brotli-1.1.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:c8146669223164fc87a7e3de9f81e9423c67a79d6b3447994dfb9c95da16e2d6", size = 446244, upload-time = "2023-09-07T14:03:39.223Z" },
-    { url = "https://files.pythonhosted.org/packages/9d/e6/f305eb61fb9a8580c525478a4a34c5ae1a9bcb12c3aee619114940bc513d/Brotli-1.1.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:30924eb4c57903d5a7526b08ef4a584acc22ab1ffa085faceb521521d2de32dd", size = 2906500, upload-time = "2023-09-07T14:03:40.858Z" },
-    { url = "https://files.pythonhosted.org/packages/3e/4f/af6846cfbc1550a3024e5d3775ede1e00474c40882c7bf5b37a43ca35e91/Brotli-1.1.0-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:ceb64bbc6eac5a140ca649003756940f8d6a7c444a68af170b3187623b43bebf", size = 2943950, upload-time = "2023-09-07T14:03:42.896Z" },
-    { url = "https://files.pythonhosted.org/packages/b3/e7/ca2993c7682d8629b62630ebf0d1f3bb3d579e667ce8e7ca03a0a0576a2d/Brotli-1.1.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a469274ad18dc0e4d316eefa616d1d0c2ff9da369af19fa6f3daa4f09671fd61", size = 2918527, upload-time = "2023-09-07T14:03:44.552Z" },
-    { url = "https://files.pythonhosted.org/packages/b3/96/da98e7bedc4c51104d29cc61e5f449a502dd3dbc211944546a4cc65500d3/Brotli-1.1.0-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:524f35912131cc2cabb00edfd8d573b07f2d9f21fa824bd3fb19725a9cf06327", size = 2845489, upload-time = "2023-09-07T14:03:46.594Z" },
-    { url = "https://files.pythonhosted.org/packages/e8/ef/ccbc16947d6ce943a7f57e1a40596c75859eeb6d279c6994eddd69615265/Brotli-1.1.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:5b3cc074004d968722f51e550b41a27be656ec48f8afaeeb45ebf65b561481dd", size = 2914080, upload-time = "2023-09-07T14:03:48.204Z" },
-    { url = "https://files.pythonhosted.org/packages/80/d6/0bd38d758d1afa62a5524172f0b18626bb2392d717ff94806f741fcd5ee9/Brotli-1.1.0-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:19c116e796420b0cee3da1ccec3b764ed2952ccfcc298b55a10e5610ad7885f9", size = 2813051, upload-time = "2023-09-07T14:03:50.348Z" },
-    { url = "https://files.pythonhosted.org/packages/14/56/48859dd5d129d7519e001f06dcfbb6e2cf6db92b2702c0c2ce7d97e086c1/Brotli-1.1.0-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:510b5b1bfbe20e1a7b3baf5fed9e9451873559a976c1a78eebaa3b86c57b4265", size = 2938172, upload-time = "2023-09-07T14:03:52.395Z" },
-    { url = "https://files.pythonhosted.org/packages/3d/77/a236d5f8cd9e9f4348da5acc75ab032ab1ab2c03cc8f430d24eea2672888/Brotli-1.1.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:a1fd8a29719ccce974d523580987b7f8229aeace506952fa9ce1d53a033873c8", size = 2933023, upload-time = "2023-09-07T14:03:53.96Z" },
-    { url = "https://files.pythonhosted.org/packages/f1/87/3b283efc0f5cb35f7f84c0c240b1e1a1003a5e47141a4881bf87c86d0ce2/Brotli-1.1.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:c247dd99d39e0338a604f8c2b3bc7061d5c2e9e2ac7ba9cc1be5a69cb6cd832f", size = 2935871, upload-time = "2024-10-18T12:32:16.688Z" },
-    { url = "https://files.pythonhosted.org/packages/f3/eb/2be4cc3e2141dc1a43ad4ca1875a72088229de38c68e842746b342667b2a/Brotli-1.1.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:1b2c248cd517c222d89e74669a4adfa5577e06ab68771a529060cf5a156e9757", size = 2847784, upload-time = "2024-10-18T12:32:18.459Z" },
-    { url = "https://files.pythonhosted.org/packages/66/13/b58ddebfd35edde572ccefe6890cf7c493f0c319aad2a5badee134b4d8ec/Brotli-1.1.0-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:2a24c50840d89ded6c9a8fdc7b6ed3692ed4e86f1c4a4a938e1e92def92933e0", size = 3034905, upload-time = "2024-10-18T12:32:20.192Z" },
-    { url = "https://files.pythonhosted.org/packages/84/9c/bc96b6c7db824998a49ed3b38e441a2cae9234da6fa11f6ed17e8cf4f147/Brotli-1.1.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:f31859074d57b4639318523d6ffdca586ace54271a73ad23ad021acd807eb14b", size = 2929467, upload-time = "2024-10-18T12:32:21.774Z" },
-    { url = "https://files.pythonhosted.org/packages/e7/71/8f161dee223c7ff7fea9d44893fba953ce97cf2c3c33f78ba260a91bcff5/Brotli-1.1.0-cp311-cp311-win32.whl", hash = "sha256:39da8adedf6942d76dc3e46653e52df937a3c4d6d18fdc94a7c29d263b1f5b50", size = 333169, upload-time = "2023-09-07T14:03:55.404Z" },
-    { url = "https://files.pythonhosted.org/packages/02/8a/fece0ee1057643cb2a5bbf59682de13f1725f8482b2c057d4e799d7ade75/Brotli-1.1.0-cp311-cp311-win_amd64.whl", hash = "sha256:aac0411d20e345dc0920bdec5548e438e999ff68d77564d5e9463a7ca9d3e7b1", size = 357253, upload-time = "2023-09-07T14:03:56.643Z" },
-    { url = "https://files.pythonhosted.org/packages/5c/d0/5373ae13b93fe00095a58efcbce837fd470ca39f703a235d2a999baadfbc/Brotli-1.1.0-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:32d95b80260d79926f5fab3c41701dbb818fde1c9da590e77e571eefd14abe28", size = 815693, upload-time = "2024-10-18T12:32:23.824Z" },
-    { url = "https://files.pythonhosted.org/packages/8e/48/f6e1cdf86751300c288c1459724bfa6917a80e30dbfc326f92cea5d3683a/Brotli-1.1.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:b760c65308ff1e462f65d69c12e4ae085cff3b332d894637f6273a12a482d09f", size = 422489, upload-time = "2024-10-18T12:32:25.641Z" },
-    { url = "https://files.pythonhosted.org/packages/06/88/564958cedce636d0f1bed313381dfc4b4e3d3f6015a63dae6146e1b8c65c/Brotli-1.1.0-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:316cc9b17edf613ac76b1f1f305d2a748f1b976b033b049a6ecdfd5612c70409", size = 873081, upload-time = "2023-09-07T14:03:57.967Z" },
-    { url = "https://files.pythonhosted.org/packages/58/79/b7026a8bb65da9a6bb7d14329fd2bd48d2b7f86d7329d5cc8ddc6a90526f/Brotli-1.1.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:caf9ee9a5775f3111642d33b86237b05808dafcd6268faa492250e9b78046eb2", size = 446244, upload-time = "2023-09-07T14:03:59.319Z" },
-    { url = "https://files.pythonhosted.org/packages/e5/18/c18c32ecea41b6c0004e15606e274006366fe19436b6adccc1ae7b2e50c2/Brotli-1.1.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:70051525001750221daa10907c77830bc889cb6d865cc0b813d9db7fefc21451", size = 2906505, upload-time = "2023-09-07T14:04:01.327Z" },
-    { url = "https://files.pythonhosted.org/packages/08/c8/69ec0496b1ada7569b62d85893d928e865df29b90736558d6c98c2031208/Brotli-1.1.0-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:7f4bf76817c14aa98cc6697ac02f3972cb8c3da93e9ef16b9c66573a68014f91", size = 2944152, upload-time = "2023-09-07T14:04:03.033Z" },
-    { url = "https://files.pythonhosted.org/packages/ab/fb/0517cea182219d6768113a38167ef6d4eb157a033178cc938033a552ed6d/Brotli-1.1.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:d0c5516f0aed654134a2fc936325cc2e642f8a0e096d075209672eb321cff408", size = 2919252, upload-time = "2023-09-07T14:04:04.675Z" },
-    { url = "https://files.pythonhosted.org/packages/c7/53/73a3431662e33ae61a5c80b1b9d2d18f58dfa910ae8dd696e57d39f1a2f5/Brotli-1.1.0-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:6c3020404e0b5eefd7c9485ccf8393cfb75ec38ce75586e046573c9dc29967a0", size = 2845955, upload-time = "2023-09-07T14:04:06.585Z" },
-    { url = "https://files.pythonhosted.org/packages/55/ac/bd280708d9c5ebdbf9de01459e625a3e3803cce0784f47d633562cf40e83/Brotli-1.1.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:4ed11165dd45ce798d99a136808a794a748d5dc38511303239d4e2363c0695dc", size = 2914304, upload-time = "2023-09-07T14:04:08.668Z" },
-    { url = "https://files.pythonhosted.org/packages/76/58/5c391b41ecfc4527d2cc3350719b02e87cb424ef8ba2023fb662f9bf743c/Brotli-1.1.0-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:4093c631e96fdd49e0377a9c167bfd75b6d0bad2ace734c6eb20b348bc3ea180", size = 2814452, upload-time = "2023-09-07T14:04:10.736Z" },
-    { url = "https://files.pythonhosted.org/packages/c7/4e/91b8256dfe99c407f174924b65a01f5305e303f486cc7a2e8a5d43c8bec3/Brotli-1.1.0-cp312-cp312-musllinux_1_1_ppc64le.whl", hash = "sha256:7e4c4629ddad63006efa0ef968c8e4751c5868ff0b1c5c40f76524e894c50248", size = 2938751, upload-time = "2023-09-07T14:04:12.875Z" },
-    { url = "https://files.pythonhosted.org/packages/5a/a6/e2a39a5d3b412938362bbbeba5af904092bf3f95b867b4a3eb856104074e/Brotli-1.1.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:861bf317735688269936f755fa136a99d1ed526883859f86e41a5d43c61d8966", size = 2933757, upload-time = "2023-09-07T14:04:14.551Z" },
-    { url = "https://files.pythonhosted.org/packages/13/f0/358354786280a509482e0e77c1a5459e439766597d280f28cb097642fc26/Brotli-1.1.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:87a3044c3a35055527ac75e419dfa9f4f3667a1e887ee80360589eb8c90aabb9", size = 2936146, upload-time = "2024-10-18T12:32:27.257Z" },
-    { url = "https://files.pythonhosted.org/packages/80/f7/daf538c1060d3a88266b80ecc1d1c98b79553b3f117a485653f17070ea2a/Brotli-1.1.0-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:c5529b34c1c9d937168297f2c1fde7ebe9ebdd5e121297ff9c043bdb2ae3d6fb", size = 2848055, upload-time = "2024-10-18T12:32:29.376Z" },
-    { url = "https://files.pythonhosted.org/packages/ad/cf/0eaa0585c4077d3c2d1edf322d8e97aabf317941d3a72d7b3ad8bce004b0/Brotli-1.1.0-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:ca63e1890ede90b2e4454f9a65135a4d387a4585ff8282bb72964fab893f2111", size = 3035102, upload-time = "2024-10-18T12:32:31.371Z" },
-    { url = "https://files.pythonhosted.org/packages/d8/63/1c1585b2aa554fe6dbce30f0c18bdbc877fa9a1bf5ff17677d9cca0ac122/Brotli-1.1.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:e79e6520141d792237c70bcd7a3b122d00f2613769ae0cb61c52e89fd3443839", size = 2930029, upload-time = "2024-10-18T12:32:33.293Z" },
-    { url = "https://files.pythonhosted.org/packages/5f/3b/4e3fd1893eb3bbfef8e5a80d4508bec17a57bb92d586c85c12d28666bb13/Brotli-1.1.0-cp312-cp312-win32.whl", hash = "sha256:5f4d5ea15c9382135076d2fb28dde923352fe02951e66935a9efaac8f10e81b0", size = 333276, upload-time = "2023-09-07T14:04:16.49Z" },
-    { url = "https://files.pythonhosted.org/packages/3d/d5/942051b45a9e883b5b6e98c041698b1eb2012d25e5948c58d6bf85b1bb43/Brotli-1.1.0-cp312-cp312-win_amd64.whl", hash = "sha256:906bc3a79de8c4ae5b86d3d75a8b77e44404b0f4261714306e3ad248d8ab0951", size = 357255, upload-time = "2023-09-07T14:04:17.83Z" },
-    { url = "https://files.pythonhosted.org/packages/0a/9f/fb37bb8ffc52a8da37b1c03c459a8cd55df7a57bdccd8831d500e994a0ca/Brotli-1.1.0-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:8bf32b98b75c13ec7cf774164172683d6e7891088f6316e54425fde1efc276d5", size = 815681, upload-time = "2024-10-18T12:32:34.942Z" },
-    { url = "https://files.pythonhosted.org/packages/06/b3/dbd332a988586fefb0aa49c779f59f47cae76855c2d00f450364bb574cac/Brotli-1.1.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:7bc37c4d6b87fb1017ea28c9508b36bbcb0c3d18b4260fcdf08b200c74a6aee8", size = 422475, upload-time = "2024-10-18T12:32:36.485Z" },
-    { url = "https://files.pythonhosted.org/packages/bb/80/6aaddc2f63dbcf2d93c2d204e49c11a9ec93a8c7c63261e2b4bd35198283/Brotli-1.1.0-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3c0ef38c7a7014ffac184db9e04debe495d317cc9c6fb10071f7fefd93100a4f", size = 2906173, upload-time = "2024-10-18T12:32:37.978Z" },
-    { url = "https://files.pythonhosted.org/packages/ea/1d/e6ca79c96ff5b641df6097d299347507d39a9604bde8915e76bf026d6c77/Brotli-1.1.0-cp313-cp313-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:91d7cc2a76b5567591d12c01f019dd7afce6ba8cba6571187e21e2fc418ae648", size = 2943803, upload-time = "2024-10-18T12:32:39.606Z" },
-    { url = "https://files.pythonhosted.org/packages/ac/a3/d98d2472e0130b7dd3acdbb7f390d478123dbf62b7d32bda5c830a96116d/Brotli-1.1.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a93dde851926f4f2678e704fadeb39e16c35d8baebd5252c9fd94ce8ce68c4a0", size = 2918946, upload-time = "2024-10-18T12:32:41.679Z" },
-    { url = "https://files.pythonhosted.org/packages/c4/a5/c69e6d272aee3e1423ed005d8915a7eaa0384c7de503da987f2d224d0721/Brotli-1.1.0-cp313-cp313-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:f0db75f47be8b8abc8d9e31bc7aad0547ca26f24a54e6fd10231d623f183d089", size = 2845707, upload-time = "2024-10-18T12:32:43.478Z" },
-    { url = "https://files.pythonhosted.org/packages/58/9f/4149d38b52725afa39067350696c09526de0125ebfbaab5acc5af28b42ea/Brotli-1.1.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:6967ced6730aed543b8673008b5a391c3b1076d834ca438bbd70635c73775368", size = 2936231, upload-time = "2024-10-18T12:32:45.224Z" },
-    { url = "https://files.pythonhosted.org/packages/5a/5a/145de884285611838a16bebfdb060c231c52b8f84dfbe52b852a15780386/Brotli-1.1.0-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:7eedaa5d036d9336c95915035fb57422054014ebdeb6f3b42eac809928e40d0c", size = 2848157, upload-time = "2024-10-18T12:32:46.894Z" },
-    { url = "https://files.pythonhosted.org/packages/50/ae/408b6bfb8525dadebd3b3dd5b19d631da4f7d46420321db44cd99dcf2f2c/Brotli-1.1.0-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:d487f5432bf35b60ed625d7e1b448e2dc855422e87469e3f450aa5552b0eb284", size = 3035122, upload-time = "2024-10-18T12:32:48.844Z" },
-    { url = "https://files.pythonhosted.org/packages/af/85/a94e5cfaa0ca449d8f91c3d6f78313ebf919a0dbd55a100c711c6e9655bc/Brotli-1.1.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:832436e59afb93e1836081a20f324cb185836c617659b07b129141a8426973c7", size = 2930206, upload-time = "2024-10-18T12:32:51.198Z" },
-    { url = "https://files.pythonhosted.org/packages/c2/f0/a61d9262cd01351df22e57ad7c34f66794709acab13f34be2675f45bf89d/Brotli-1.1.0-cp313-cp313-win32.whl", hash = "sha256:43395e90523f9c23a3d5bdf004733246fba087f2948f87ab28015f12359ca6a0", size = 333804, upload-time = "2024-10-18T12:32:52.661Z" },
-    { url = "https://files.pythonhosted.org/packages/7e/c1/ec214e9c94000d1c1974ec67ced1c970c148aa6b8d8373066123fc3dbf06/Brotli-1.1.0-cp313-cp313-win_amd64.whl", hash = "sha256:9011560a466d2eb3f5a6e4929cf4a09be405c64154e12df0dd72713f6500e32b", size = 358517, upload-time = "2024-10-18T12:32:54.066Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/ef/f285668811a9e1ddb47a18cb0b437d5fc2760d537a2fe8a57875ad6f8448/brotli-1.2.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:15b33fe93cedc4caaff8a0bd1eb7e3dab1c61bb22a0bf5bdfdfd97cd7da79744", size = 863110, upload-time = "2025-11-05T18:38:12.978Z" },
+    { url = "https://files.pythonhosted.org/packages/50/62/a3b77593587010c789a9d6eaa527c79e0848b7b860402cc64bc0bc28a86c/brotli-1.2.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:898be2be399c221d2671d29eed26b6b2713a02c2119168ed914e7d00ceadb56f", size = 445438, upload-time = "2025-11-05T18:38:14.208Z" },
+    { url = "https://files.pythonhosted.org/packages/cd/e1/7fadd47f40ce5549dc44493877db40292277db373da5053aff181656e16e/brotli-1.2.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:350c8348f0e76fff0a0fd6c26755d2653863279d086d3aa2c290a6a7251135dd", size = 1534420, upload-time = "2025-11-05T18:38:15.111Z" },
+    { url = "https://files.pythonhosted.org/packages/12/8b/1ed2f64054a5a008a4ccd2f271dbba7a5fb1a3067a99f5ceadedd4c1d5a7/brotli-1.2.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:2e1ad3fda65ae0d93fec742a128d72e145c9c7a99ee2fcd667785d99eb25a7fe", size = 1632619, upload-time = "2025-11-05T18:38:16.094Z" },
+    { url = "https://files.pythonhosted.org/packages/89/5a/7071a621eb2d052d64efd5da2ef55ecdac7c3b0c6e4f9d519e9c66d987ef/brotli-1.2.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:40d918bce2b427a0c4ba189df7a006ac0c7277c180aee4617d99e9ccaaf59e6a", size = 1426014, upload-time = "2025-11-05T18:38:17.177Z" },
+    { url = "https://files.pythonhosted.org/packages/26/6d/0971a8ea435af5156acaaccec1a505f981c9c80227633851f2810abd252a/brotli-1.2.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:2a7f1d03727130fc875448b65b127a9ec5d06d19d0148e7554384229706f9d1b", size = 1489661, upload-time = "2025-11-05T18:38:18.41Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/75/c1baca8b4ec6c96a03ef8230fab2a785e35297632f402ebb1e78a1e39116/brotli-1.2.0-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:9c79f57faa25d97900bfb119480806d783fba83cd09ee0b33c17623935b05fa3", size = 1599150, upload-time = "2025-11-05T18:38:19.792Z" },
+    { url = "https://files.pythonhosted.org/packages/0d/1a/23fcfee1c324fd48a63d7ebf4bac3a4115bdb1b00e600f80f727d850b1ae/brotli-1.2.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:844a8ceb8483fefafc412f85c14f2aae2fb69567bf2a0de53cdb88b73e7c43ae", size = 1493505, upload-time = "2025-11-05T18:38:20.913Z" },
+    { url = "https://files.pythonhosted.org/packages/36/e5/12904bbd36afeef53d45a84881a4810ae8810ad7e328a971ebbfd760a0b3/brotli-1.2.0-cp311-cp311-win32.whl", hash = "sha256:aa47441fa3026543513139cb8926a92a8e305ee9c71a6209ef7a97d91640ea03", size = 334451, upload-time = "2025-11-05T18:38:21.94Z" },
+    { url = "https://files.pythonhosted.org/packages/02/8b/ecb5761b989629a4758c394b9301607a5880de61ee2ee5fe104b87149ebc/brotli-1.2.0-cp311-cp311-win_amd64.whl", hash = "sha256:022426c9e99fd65d9475dce5c195526f04bb8be8907607e27e747893f6ee3e24", size = 369035, upload-time = "2025-11-05T18:38:22.941Z" },
+    { url = "https://files.pythonhosted.org/packages/11/ee/b0a11ab2315c69bb9b45a2aaed022499c9c24a205c3a49c3513b541a7967/brotli-1.2.0-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:35d382625778834a7f3061b15423919aa03e4f5da34ac8e02c074e4b75ab4f84", size = 861543, upload-time = "2025-11-05T18:38:24.183Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/2f/29c1459513cd35828e25531ebfcbf3e92a5e49f560b1777a9af7203eb46e/brotli-1.2.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:7a61c06b334bd99bc5ae84f1eeb36bfe01400264b3c352f968c6e30a10f9d08b", size = 444288, upload-time = "2025-11-05T18:38:25.139Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/6f/feba03130d5fceadfa3a1bb102cb14650798c848b1df2a808356f939bb16/brotli-1.2.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:acec55bb7c90f1dfc476126f9711a8e81c9af7fb617409a9ee2953115343f08d", size = 1528071, upload-time = "2025-11-05T18:38:26.081Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/38/f3abb554eee089bd15471057ba85f47e53a44a462cfce265d9bf7088eb09/brotli-1.2.0-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:260d3692396e1895c5034f204f0db022c056f9e2ac841593a4cf9426e2a3faca", size = 1626913, upload-time = "2025-11-05T18:38:27.284Z" },
+    { url = "https://files.pythonhosted.org/packages/03/a7/03aa61fbc3c5cbf99b44d158665f9b0dd3d8059be16c460208d9e385c837/brotli-1.2.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:072e7624b1fc4d601036ab3f4f27942ef772887e876beff0301d261210bca97f", size = 1419762, upload-time = "2025-11-05T18:38:28.295Z" },
+    { url = "https://files.pythonhosted.org/packages/21/1b/0374a89ee27d152a5069c356c96b93afd1b94eae83f1e004b57eb6ce2f10/brotli-1.2.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:adedc4a67e15327dfdd04884873c6d5a01d3e3b6f61406f99b1ed4865a2f6d28", size = 1484494, upload-time = "2025-11-05T18:38:29.29Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/57/69d4fe84a67aef4f524dcd075c6eee868d7850e85bf01d778a857d8dbe0a/brotli-1.2.0-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:7a47ce5c2288702e09dc22a44d0ee6152f2c7eda97b3c8482d826a1f3cfc7da7", size = 1593302, upload-time = "2025-11-05T18:38:30.639Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/3b/39e13ce78a8e9a621c5df3aeb5fd181fcc8caba8c48a194cd629771f6828/brotli-1.2.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:af43b8711a8264bb4e7d6d9a6d004c3a2019c04c01127a868709ec29962b6036", size = 1487913, upload-time = "2025-11-05T18:38:31.618Z" },
+    { url = "https://files.pythonhosted.org/packages/62/28/4d00cb9bd76a6357a66fcd54b4b6d70288385584063f4b07884c1e7286ac/brotli-1.2.0-cp312-cp312-win32.whl", hash = "sha256:e99befa0b48f3cd293dafeacdd0d191804d105d279e0b387a32054c1180f3161", size = 334362, upload-time = "2025-11-05T18:38:32.939Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/4e/bc1dcac9498859d5e353c9b153627a3752868a9d5f05ce8dedd81a2354ab/brotli-1.2.0-cp312-cp312-win_amd64.whl", hash = "sha256:b35c13ce241abdd44cb8ca70683f20c0c079728a36a996297adb5334adfc1c44", size = 369115, upload-time = "2025-11-05T18:38:33.765Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/d4/4ad5432ac98c73096159d9ce7ffeb82d151c2ac84adcc6168e476bb54674/brotli-1.2.0-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:9e5825ba2c9998375530504578fd4d5d1059d09621a02065d1b6bfc41a8e05ab", size = 861523, upload-time = "2025-11-05T18:38:34.67Z" },
+    { url = "https://files.pythonhosted.org/packages/91/9f/9cc5bd03ee68a85dc4bc89114f7067c056a3c14b3d95f171918c088bf88d/brotli-1.2.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:0cf8c3b8ba93d496b2fae778039e2f5ecc7cff99df84df337ca31d8f2252896c", size = 444289, upload-time = "2025-11-05T18:38:35.6Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/b6/fe84227c56a865d16a6614e2c4722864b380cb14b13f3e6bef441e73a85a/brotli-1.2.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:c8565e3cdc1808b1a34714b553b262c5de5fbda202285782173ec137fd13709f", size = 1528076, upload-time = "2025-11-05T18:38:36.639Z" },
+    { url = "https://files.pythonhosted.org/packages/55/de/de4ae0aaca06c790371cf6e7ee93a024f6b4bb0568727da8c3de112e726c/brotli-1.2.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:26e8d3ecb0ee458a9804f47f21b74845cc823fd1bb19f02272be70774f56e2a6", size = 1626880, upload-time = "2025-11-05T18:38:37.623Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/16/a1b22cbea436642e071adcaf8d4b350a2ad02f5e0ad0da879a1be16188a0/brotli-1.2.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:67a91c5187e1eec76a61625c77a6c8c785650f5b576ca732bd33ef58b0dff49c", size = 1419737, upload-time = "2025-11-05T18:38:38.729Z" },
+    { url = "https://files.pythonhosted.org/packages/46/63/c968a97cbb3bdbf7f974ef5a6ab467a2879b82afbc5ffb65b8acbb744f95/brotli-1.2.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:4ecdb3b6dc36e6d6e14d3a1bdc6c1057c8cbf80db04031d566eb6080ce283a48", size = 1484440, upload-time = "2025-11-05T18:38:39.916Z" },
+    { url = "https://files.pythonhosted.org/packages/06/9d/102c67ea5c9fc171f423e8399e585dabea29b5bc79b05572891e70013cdd/brotli-1.2.0-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:3e1b35d56856f3ed326b140d3c6d9db91740f22e14b06e840fe4bb1923439a18", size = 1593313, upload-time = "2025-11-05T18:38:41.24Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/4a/9526d14fa6b87bc827ba1755a8440e214ff90de03095cacd78a64abe2b7d/brotli-1.2.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:54a50a9dad16b32136b2241ddea9e4df159b41247b2ce6aac0b3276a66a8f1e5", size = 1487945, upload-time = "2025-11-05T18:38:42.277Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/e8/3fe1ffed70cbef83c5236166acaed7bb9c766509b157854c80e2f766b38c/brotli-1.2.0-cp313-cp313-win32.whl", hash = "sha256:1b1d6a4efedd53671c793be6dd760fcf2107da3a52331ad9ea429edf0902f27a", size = 334368, upload-time = "2025-11-05T18:38:43.345Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/91/e739587be970a113b37b821eae8097aac5a48e5f0eca438c22e4c7dd8648/brotli-1.2.0-cp313-cp313-win_amd64.whl", hash = "sha256:b63daa43d82f0cdabf98dee215b375b4058cce72871fd07934f179885aad16e8", size = 369116, upload-time = "2025-11-05T18:38:44.609Z" },
+    { url = "https://files.pythonhosted.org/packages/17/e1/298c2ddf786bb7347a1cd71d63a347a79e5712a7c0cba9e3c3458ebd976f/brotli-1.2.0-cp314-cp314-macosx_10_15_universal2.whl", hash = "sha256:6c12dad5cd04530323e723787ff762bac749a7b256a5bece32b2243dd5c27b21", size = 863080, upload-time = "2025-11-05T18:38:45.503Z" },
+    { url = "https://files.pythonhosted.org/packages/84/0c/aac98e286ba66868b2b3b50338ffbd85a35c7122e9531a73a37a29763d38/brotli-1.2.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:3219bd9e69868e57183316ee19c84e03e8f8b5a1d1f2667e1aa8c2f91cb061ac", size = 445453, upload-time = "2025-11-05T18:38:46.433Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/f1/0ca1f3f99ae300372635ab3fe2f7a79fa335fee3d874fa7f9e68575e0e62/brotli-1.2.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:963a08f3bebd8b75ac57661045402da15991468a621f014be54e50f53a58d19e", size = 1528168, upload-time = "2025-11-05T18:38:47.371Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/a6/2ebfc8f766d46df8d3e65b880a2e220732395e6d7dc312c1e1244b0f074a/brotli-1.2.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:9322b9f8656782414b37e6af884146869d46ab85158201d82bab9abbcb971dc7", size = 1627098, upload-time = "2025-11-05T18:38:48.385Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/2f/0976d5b097ff8a22163b10617f76b2557f15f0f39d6a0fe1f02b1a53e92b/brotli-1.2.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:cf9cba6f5b78a2071ec6fb1e7bd39acf35071d90a81231d67e92d637776a6a63", size = 1419861, upload-time = "2025-11-05T18:38:49.372Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/97/d76df7176a2ce7616ff94c1fb72d307c9a30d2189fe877f3dd99af00ea5a/brotli-1.2.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:7547369c4392b47d30a3467fe8c3330b4f2e0f7730e45e3103d7d636678a808b", size = 1484594, upload-time = "2025-11-05T18:38:50.655Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/93/14cf0b1216f43df5609f5b272050b0abd219e0b54ea80b47cef9867b45e7/brotli-1.2.0-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:fc1530af5c3c275b8524f2e24841cbe2599d74462455e9bae5109e9ff42e9361", size = 1593455, upload-time = "2025-11-05T18:38:51.624Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/73/3183c9e41ca755713bdf2cc1d0810df742c09484e2e1ddd693bee53877c1/brotli-1.2.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:d2d085ded05278d1c7f65560aae97b3160aeb2ea2c0b3e26204856beccb60888", size = 1488164, upload-time = "2025-11-05T18:38:53.079Z" },
+    { url = "https://files.pythonhosted.org/packages/64/6a/0c78d8f3a582859236482fd9fa86a65a60328a00983006bcf6d83b7b2253/brotli-1.2.0-cp314-cp314-win32.whl", hash = "sha256:832c115a020e463c2f67664560449a7bea26b0c1fdd690352addad6d0a08714d", size = 339280, upload-time = "2025-11-05T18:38:54.02Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/10/56978295c14794b2c12007b07f3e41ba26acda9257457d7085b0bb3bb90c/brotli-1.2.0-cp314-cp314-win_amd64.whl", hash = "sha256:e7c0af964e0b4e3412a0ebf341ea26ec767fa0b4cf81abb5e897c9338b5ad6a3", size = 375639, upload-time = "2025-11-05T18:38:55.67Z" },
 ]
 
 [[package]]
@@ -364,32 +408,39 @@ wheels = [
 
 [[package]]
 name = "cbor2"
-version = "5.6.5"
+version = "5.9.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e4/aa/ba55b47d51d27911981a18743b4d3cebfabccbb0598c09801b734cec4184/cbor2-5.6.5.tar.gz", hash = "sha256:b682820677ee1dbba45f7da11898d2720f92e06be36acec290867d5ebf3d7e09", size = 100886, upload-time = "2024-10-09T12:26:24.106Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/bd/cb/09939728be094d155b5d4ac262e39877875f5f7e36eea66beb359f647bd0/cbor2-5.9.0.tar.gz", hash = "sha256:85c7a46279ac8f226e1059275221e6b3d0e370d2bb6bd0500f9780781615bcea", size = 111231, upload-time = "2026-03-22T15:56:50.638Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/bd/b7/ef045245180510305648fd604244d3bb1ecf1b20de68f42ab5bc20198024/cbor2-5.6.5-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:863e0983989d56d5071270790e7ed8ddbda88c9e5288efdb759aba2efee670bc", size = 66452, upload-time = "2024-10-09T12:25:36.676Z" },
-    { url = "https://files.pythonhosted.org/packages/41/20/5a9d93f86b1e8fd9d9db33aff39c0e3a8459e0803ec24bd837d8b56d4a1d/cbor2-5.6.5-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:5cff06464b8f4ca6eb9abcba67bda8f8334a058abc01005c8e616728c387ad32", size = 67421, upload-time = "2024-10-09T12:25:38.114Z" },
-    { url = "https://files.pythonhosted.org/packages/0f/1e/2010f6d02dd117df88df64baf3eeca6aa6614cc81bdd6bfabf615889cf1f/cbor2-5.6.5-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f4c7dbcdc59ea7f5a745d3e30ee5e6b6ff5ce7ac244aa3de6786391b10027bb3", size = 260756, upload-time = "2024-10-09T12:25:39.657Z" },
-    { url = "https://files.pythonhosted.org/packages/e1/84/e177d9bef4749d14f31c513b25e341ac84e403e2ffa2bde562eac9e6184b/cbor2-5.6.5-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:34cf5ab0dc310c3d0196caa6ae062dc09f6c242e2544bea01691fe60c0230596", size = 249210, upload-time = "2024-10-09T12:25:41.316Z" },
-    { url = "https://files.pythonhosted.org/packages/38/75/ebfdbb281104b46419fe7cb65979de9927b75acebcb6afa0af291f728cd2/cbor2-5.6.5-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:6797b824b26a30794f2b169c0575301ca9b74ae99064e71d16e6ba0c9057de51", size = 249138, upload-time = "2024-10-09T12:25:42.432Z" },
-    { url = "https://files.pythonhosted.org/packages/b2/1e/12d887fb1a8227a16181eeec5d43057e251204626d73e1c20a77046ac1b1/cbor2-5.6.5-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:73b9647eed1493097db6aad61e03d8f1252080ee041a1755de18000dd2c05f37", size = 247156, upload-time = "2024-10-09T12:25:43.588Z" },
-    { url = "https://files.pythonhosted.org/packages/6f/76/478c12193de9517ce691bb8a3f7c00eafdd6a1bc3f7f23282ecdd99d02ec/cbor2-5.6.5-cp311-cp311-win_amd64.whl", hash = "sha256:6e14a1bf6269d25e02ef1d4008e0ce8880aa271d7c6b4c329dba48645764f60e", size = 66319, upload-time = "2024-10-09T12:25:44.621Z" },
-    { url = "https://files.pythonhosted.org/packages/57/af/84ced14c541451696825b7b8ccbb7668f688372ad8ee74aaca4311e79672/cbor2-5.6.5-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:e25c2aebc9db99af7190e2261168cdde8ed3d639ca06868e4f477cf3a228a8e9", size = 67553, upload-time = "2024-10-09T12:25:45.767Z" },
-    { url = "https://files.pythonhosted.org/packages/f2/d6/f63a840c68fed4de67d5441947af2dc695152cc488bb0e57312832fb923a/cbor2-5.6.5-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:fde21ac1cf29336a31615a2c469a9cb03cf0add3ae480672d4d38cda467d07fc", size = 67569, upload-time = "2024-10-09T12:25:46.665Z" },
-    { url = "https://files.pythonhosted.org/packages/77/ac/5fb79db6e882ec29680f4a974d35c098020a1b4709cad077667a8c3f4676/cbor2-5.6.5-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a8947c102cac79d049eadbd5e2ffb8189952890df7cbc3ee262bbc2f95b011a9", size = 276610, upload-time = "2024-10-09T12:25:48.14Z" },
-    { url = "https://files.pythonhosted.org/packages/cf/cb/70751377d94112001d46c311b5c40b45f34863dfa78a6bc71b71f40c8c7f/cbor2-5.6.5-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:38886c41bebcd7dca57739439455bce759f1e4c551b511f618b8e9c1295b431b", size = 270004, upload-time = "2024-10-09T12:25:49.769Z" },
-    { url = "https://files.pythonhosted.org/packages/f1/90/08800367e920aef31b93bd7b0cd6fadcb3a3f2243f4ed77a0d1c76f22b99/cbor2-5.6.5-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:ae2b49226224e92851c333b91d83292ec62eba53a19c68a79890ce35f1230d70", size = 264913, upload-time = "2024-10-09T12:25:50.92Z" },
-    { url = "https://files.pythonhosted.org/packages/a8/9c/76b11a5ea7548bccb0dfef3e8fb3ede48bfeb39348f0c217519e0c40d33a/cbor2-5.6.5-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:f2764804ffb6553283fc4afb10a280715905a4cea4d6dc7c90d3e89c4a93bc8d", size = 266751, upload-time = "2024-10-09T12:25:52.777Z" },
-    { url = "https://files.pythonhosted.org/packages/10/18/3866693a87c90cb12f7942e791d0f03a40ba44887dde7b7fc85319647efe/cbor2-5.6.5-cp312-cp312-win_amd64.whl", hash = "sha256:a3ac50485cf67dfaab170a3e7b527630e93cb0a6af8cdaa403054215dff93adf", size = 66739, upload-time = "2024-10-09T12:25:54.606Z" },
-    { url = "https://files.pythonhosted.org/packages/2b/69/77e93caae71d1baee927c9762e702c464715d88073133052c74ecc9d37d4/cbor2-5.6.5-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:f0d0a9c5aabd48ecb17acf56004a7542a0b8d8212be52f3102b8218284bd881e", size = 67647, upload-time = "2024-10-09T12:25:55.637Z" },
-    { url = "https://files.pythonhosted.org/packages/84/83/cb941d4fd10e4696b2c0f6fb2e3056d9a296e5765b2000a69e29a507f819/cbor2-5.6.5-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:61ceb77e6aa25c11c814d4fe8ec9e3bac0094a1f5bd8a2a8c95694596ea01e08", size = 67657, upload-time = "2024-10-09T12:25:56.528Z" },
-    { url = "https://files.pythonhosted.org/packages/5c/3f/e16a1e29994483c751b714cdf61d2956290b0b30e94690fa714a9f155c5c/cbor2-5.6.5-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:97a7e409b864fecf68b2ace8978eb5df1738799a333ec3ea2b9597bfcdd6d7d2", size = 275863, upload-time = "2024-10-09T12:25:57.462Z" },
-    { url = "https://files.pythonhosted.org/packages/64/04/f64bda3eea649fe6644c59f13d0e1f4666d975ce305cadf13835233b2a26/cbor2-5.6.5-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7f6d69f38f7d788b04c09ef2b06747536624b452b3c8b371ab78ad43b0296fab", size = 269131, upload-time = "2024-10-09T12:25:59.635Z" },
-    { url = "https://files.pythonhosted.org/packages/f4/8d/0d5ad3467f70578b032b3f52eb0f01f0327d5ae6b1f9e7d4d4e01a73aa95/cbor2-5.6.5-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:f91e6d74fa6917df31f8757fdd0e154203b0dd0609ec53eb957016a2b474896a", size = 264728, upload-time = "2024-10-09T12:26:01.407Z" },
-    { url = "https://files.pythonhosted.org/packages/77/cb/9b4f7890325eaa374c21fcccfee61a099ccb9ea0bc0f606acf7495f9568c/cbor2-5.6.5-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:5ce13a27ef8fddf643fc17a753fe34aa72b251d03c23da6a560c005dc171085b", size = 266314, upload-time = "2024-10-09T12:26:02.451Z" },
-    { url = "https://files.pythonhosted.org/packages/a8/cd/793dc041395609f5dd1edfdf0aecde504dc0fd35ed67eb3b2db79fb8ef4d/cbor2-5.6.5-cp313-cp313-win_amd64.whl", hash = "sha256:54c72a3207bb2d4480c2c39dad12d7971ce0853a99e3f9b8d559ce6eac84f66f", size = 66792, upload-time = "2024-10-09T12:26:03.615Z" },
-    { url = "https://files.pythonhosted.org/packages/9b/ef/1c4698cac96d792005ef0611832f38eaee477c275ab4b02cbfc4daba7ad3/cbor2-5.6.5-py3-none-any.whl", hash = "sha256:3038523b8fc7de312bb9cdcbbbd599987e64307c4db357cd2030c472a6c7d468", size = 23752, upload-time = "2024-10-09T12:26:23.167Z" },
+    { url = "https://files.pythonhosted.org/packages/43/aa/317c7118b8dda4c9563125c1a12c70c5b41e36677964a49c72b1aac061ec/cbor2-5.9.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:0485d3372fc832c5e16d4eb45fa1a20fc53e806e6c29a1d2b0d3e176cedd52b9", size = 70578, upload-time = "2026-03-22T15:56:03.835Z" },
+    { url = "https://files.pythonhosted.org/packages/31/43/fe29b1f897770011a5e7497f4523c2712282ee4a6cbf775ea6383fb7afb9/cbor2-5.9.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:a9d6e4e0f988b0e766509a8071975a8ee99f930e14a524620bf38083106158d2", size = 268738, upload-time = "2026-03-22T15:56:05.222Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/1a/e494568f3d8aafbcdfe361df44c3bcf5cdab5183e25ea08e3d3f9fcf4075/cbor2-5.9.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:5326336f633cc89dfe543c78829c16c3a6449c2c03277d1ddba99086c3323363", size = 262571, upload-time = "2026-03-22T15:56:06.411Z" },
+    { url = "https://files.pythonhosted.org/packages/42/2e/92acd6f87382fd44a34d9d7e85cc45372e6ba664040b72d1d9df648b25d0/cbor2-5.9.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:5e702b02d42a5ace45425b595ffe70fe35aebaf9a3cdfdc2c758b6189c744422", size = 262356, upload-time = "2026-03-22T15:56:08.236Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/68/52c039a28688baeeb78b0be7483855e6c66ea05884a937444deede0c87b8/cbor2-5.9.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:2372d357d403e7912f104ff085950ffc82a5854d6d717f1ca1ce16a40a0ef5a7", size = 257604, upload-time = "2026-03-22T15:56:09.835Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/e4/10d96a7f73ed9227090ce6e3df5d73329eb6a267dab7d5b989e6fbf6c504/cbor2-5.9.0-cp311-cp311-win_amd64.whl", hash = "sha256:1d02b65f070fd726bdc310d927228975bb655d155bf059b6eb7cacefb3dca86f", size = 69388, upload-time = "2026-03-22T15:56:11.28Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/c6/eea5829aa5a649db540f47ea35f4bf2313383d28246f0cbc50432cfad6b3/cbor2-5.9.0-cp311-cp311-win_arm64.whl", hash = "sha256:837754ece9052b3f607047e1741e5f852a538aa2b0ee3db11c82a8fa11804aa4", size = 65315, upload-time = "2026-03-22T15:56:12.326Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/39/72d8a5a4b06565561ec28f4fcb41aff7bb77f51705c01f00b8254a2aca4f/cbor2-5.9.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:1f223dffb1bcdd2764665f04c1152943d9daa4bc124a576cd8dee1cad4264313", size = 71223, upload-time = "2026-03-22T15:56:13.68Z" },
+    { url = "https://files.pythonhosted.org/packages/09/fd/7ddf3d3153b54c69c3be77172b8d9aa3a9d74f62a7fbde614d53eaeed9a4/cbor2-5.9.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ae6c706ac1d85a0b3cb3395308fd0c4d55e3202b4760773675957e93cdff45fc", size = 287865, upload-time = "2026-03-22T15:56:14.813Z" },
+    { url = "https://files.pythonhosted.org/packages/db/9d/7ede2cc42f9bb4260492e7d29d2aab781eacbbcfb09d983de1e695077199/cbor2-5.9.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4cd43d8fc374b31643b2830910f28177a606a7bc84975a62675dd3f2e320fc7b", size = 288246, upload-time = "2026-03-22T15:56:16.113Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/9d/588ebc7c5bc5843f609b05fe07be8575c7dec987735b0bbc908ac9c1264a/cbor2-5.9.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:4aa07b392cc3d76fb31c08a46a226b58c320d1c172ff3073e864409ced7bc50f", size = 280214, upload-time = "2026-03-22T15:56:17.519Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/a1/6fc8f4b15c6a27e7fbb7966c30c2b4b18c274a3221fa2f5e6235502d34bc/cbor2-5.9.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:971d425b3a23b75953d8853d5f9911bdeefa09d759ee3b5e6b07b5ff3cbd9073", size = 282162, upload-time = "2026-03-22T15:56:18.975Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/20/9a22cfe08be16ddfeef2542cf4eeed1b29f3f57ddbba0b42f7e0bb8331fd/cbor2-5.9.0-cp312-cp312-win_amd64.whl", hash = "sha256:34a6cb15e6ab6a8eae94ad2041731cd3ef786af43a8df99f847969af5b902ee7", size = 70049, upload-time = "2026-03-22T15:56:20.502Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/9e/695f92d09006614034e25a9f5b10620f3b219f79c1bec3c37b7c6f27a7a9/cbor2-5.9.0-cp312-cp312-win_arm64.whl", hash = "sha256:7d1ddc4541e7367ac58c2470cc0df847f7137167fe4f5729e2d3cc0b993d7da4", size = 65382, upload-time = "2026-03-22T15:56:21.526Z" },
+    { url = "https://files.pythonhosted.org/packages/81/c5/4901e21a8afe9448fd947b11e8f383903207cd6dd0800e5f5a386838de5b/cbor2-5.9.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:fbb06f34aa645b4deca66643bba3d400d20c15312d1fe88d429be60c1ab50f27", size = 71284, upload-time = "2026-03-22T15:56:22.836Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/10/df643a381aebc3f05486de4813662bc58accb640fc3275cb276a75e89694/cbor2-5.9.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ac684fe195c39821fca70d18afbf748f728aefbfbf88456018d299e559b8cae0", size = 287682, upload-time = "2026-03-22T15:56:24.024Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/0c/8aa6b766059ae4a0ca1ec3ff96fe3823a69a7be880dba2e249f7fbe2700b/cbor2-5.9.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:2a54fbb32cb828c214f7f333a707e4aec61182e7efdc06ea5d9596d3ecee624a", size = 288009, upload-time = "2026-03-22T15:56:25.305Z" },
+    { url = "https://files.pythonhosted.org/packages/74/07/6236bc25c183a9cf7e8062e5dddf9eae9b0b14ebf14a58a69fe5a1e872c6/cbor2-5.9.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:4753a6d1bc71054d9179557bc65740860f185095ccb401d46637fff028a5b3ec", size = 280437, upload-time = "2026-03-22T15:56:26.479Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/0a/84328d23c3c68874ac6497edb9b1900579a1028efa54734df3f1762bbc15/cbor2-5.9.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:380e534482b843e43442b87d8777a7bf9bed20cb7526f89b780c3400f617304b", size = 282247, upload-time = "2026-03-22T15:56:28.644Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/f6/89b4627e09d028c8e5fcaf7cb55f225c33ce6e037ec1844e65d02bcfa945/cbor2-5.9.0-cp313-cp313-win_amd64.whl", hash = "sha256:dcf0f695873e5c94bd072d6af8698e72b8fb7f7a18f37e0bced1041b7111a6cf", size = 70089, upload-time = "2026-03-22T15:56:29.801Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/7c/efadcd5f0102db692490e4e206988a2f98d39a09912090db497a2b800885/cbor2-5.9.0-cp313-cp313-win_arm64.whl", hash = "sha256:f7c9751a9611601ab326d8f5837f01379195bbf06175fb4effeb552140e7c9e8", size = 65466, upload-time = "2026-03-22T15:56:30.823Z" },
+    { url = "https://files.pythonhosted.org/packages/08/7d/9ccc36d10ef96e6038e48046ebe1ce35a1e7814da0e1e204d09e6ef09b8d/cbor2-5.9.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:23606d31ba1368bd1b6602e3020ee88fe9523ca80e8630faf6b2fc904fd84560", size = 71500, upload-time = "2026-03-22T15:56:31.876Z" },
+    { url = "https://files.pythonhosted.org/packages/70/e1/a6cca2cc72e13f00030c6a649f57ae703eb2c620806ab70c40db8eab33fa/cbor2-5.9.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0322296b9d52f55880e300ba8ba09ecf644303b99b51138bbb1c0fb644fa7c3e", size = 286953, upload-time = "2026-03-22T15:56:33.292Z" },
+    { url = "https://files.pythonhosted.org/packages/08/3c/24cd5ef488a957d90e016f200a3aad820e4c2f85edd61c9fe4523007a1ee/cbor2-5.9.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:422817286c1d0ce947fb2f7eca9212b39bddd7231e8b452e2d2cc52f15332dba", size = 285454, upload-time = "2026-03-22T15:56:34.703Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/35/dca96818494c0ba47cdd73e8d809b27fa91f8fa0ce32a068a09237687454/cbor2-5.9.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:9a4907e0c3035bb8836116854ed8e56d8aef23909d601fa59706320897ec2551", size = 279441, upload-time = "2026-03-22T15:56:35.888Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/44/d3362378b16e53cf7e535a3f5aed8476e2109068154e24e31981ef5bde9e/cbor2-5.9.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:fb7afe77f8d269e42d7c4b515c6fd14f1ccc0625379fb6829b269f493d16eddd", size = 279673, upload-time = "2026-03-22T15:56:37.08Z" },
+    { url = "https://files.pythonhosted.org/packages/43/d1/3533a697e5842fff7c2f64912eb251f8dcab3a8b5d88e228d6eebc3b5021/cbor2-5.9.0-cp314-cp314-win_amd64.whl", hash = "sha256:86baf870d4c0bfc6f79de3801f3860a84ab76d9c8b0abb7f081f2c14c38d79d3", size = 71940, upload-time = "2026-03-22T15:56:38.366Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/e2/c6ba75f3fb25dfa15ab6999cc8709c821987e9ed8e375d7f58539261bcb9/cbor2-5.9.0-cp314-cp314-win_arm64.whl", hash = "sha256:7221483fad0c63afa4244624d552abf89d7dfdbc5f5edfc56fc1ff2b4b818975", size = 67639, upload-time = "2026-03-22T15:56:39.39Z" },
+    { url = "https://files.pythonhosted.org/packages/42/ff/b83492b096fbef26e9cb62c1a4bf2d3cef579ea7b33138c6c37c4ae66f67/cbor2-5.9.0-py3-none-any.whl", hash = "sha256:27695cbd70c90b8de5c4a284642c2836449b14e2c2e07e3ffe0744cb7669a01b", size = 24627, upload-time = "2026-03-22T15:56:48.847Z" },
 ]
 
 [[package]]
@@ -1475,14 +1526,14 @@ wheels = [
 
 [[package]]
 name = "marshmallow"
-version = "3.26.1"
+version = "3.26.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "packaging" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/ab/5e/5e53d26b42ab75491cda89b871dab9e97c840bf12c63ec58a1919710cd06/marshmallow-3.26.1.tar.gz", hash = "sha256:e6d8affb6cb61d39d26402096dc0aee12d5a26d490a121f118d2e81dc0719dc6", size = 221825, upload-time = "2025-02-03T15:32:25.093Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/55/79/de6c16cc902f4fc372236926b0ce2ab7845268dcc30fb2fbb7f71b418631/marshmallow-3.26.2.tar.gz", hash = "sha256:bbe2adb5a03e6e3571b573f42527c6fe926e17467833660bebd11593ab8dfd57", size = 222095, upload-time = "2025-12-22T06:53:53.309Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/34/75/51952c7b2d3873b44a0028b1bd26a25078c18f92f256608e8d1dc61b39fd/marshmallow-3.26.1-py3-none-any.whl", hash = "sha256:3350409f20a70a7e4e11a27661187b77cdcaeb20abca41c1454fe33636bea09c", size = 50878, upload-time = "2025-02-03T15:32:22.295Z" },
+    { url = "https://files.pythonhosted.org/packages/be/2f/5108cb3ee4ba6501748c4908b908e55f42a5b66245b4cfe0c99326e1ef6e/marshmallow-3.26.2-py3-none-any.whl", hash = "sha256:013fa8a3c4c276c24d26d84ce934dc964e2aa794345a0f8c7e5a7191482c8a73", size = 50964, upload-time = "2025-12-22T06:53:51.801Z" },
 ]
 
 [[package]]
@@ -2062,16 +2113,16 @@ wheels = [
 
 [[package]]
 name = "pydantic-settings"
-version = "2.14.0"
+version = "2.14.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "pydantic" },
     { name = "python-dotenv" },
     { name = "typing-inspection" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/42/98/c8345dccdc31de4228c039a98f6467a941e39558da41c1744fbe29fa5666/pydantic_settings-2.14.0.tar.gz", hash = "sha256:24285fd4b0e0c06507dd9fdfd331ee23794305352aaec8fc4eb92d4047aeb67d", size = 235709, upload-time = "2026-04-20T13:37:40.293Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/5c/b5/8f48e906c3e0205276e8bd8cb7512217a87b2685304d64be27cad5b3019f/pydantic_settings-2.14.2.tar.gz", hash = "sha256:c19dd64b19097f1de80184f0cc7b0272a13ae6e170cbf240a3e27e381ed14a5f", size = 237700, upload-time = "2026-06-19T13:44:56.324Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/01/dd/bebff3040138f00ae8a102d426b27349b9a49acc310fcae7f92112d867e3/pydantic_settings-2.14.0-py3-none-any.whl", hash = "sha256:fc8d5d692eb7092e43c8647c1c35a3ecd00e040fcf02ed86f4cb5458ca62182e", size = 60940, upload-time = "2026-04-20T13:37:38.586Z" },
+    { url = "https://files.pythonhosted.org/packages/77/c1/6e422f34e569cf8e18df68d1939c81c099d2b61e4f7d9621c8a77560799c/pydantic_settings-2.14.2-py3-none-any.whl", hash = "sha256:a20c97b37910b6550d5ea50fbcc2d4187defe58cd57070b73863d069419c9440", size = 61715, upload-time = "2026-06-19T13:44:55.02Z" },
 ]
 
 [[package]]
@@ -2085,11 +2136,11 @@ wheels = [
 
 [[package]]
 name = "pyjwt"
-version = "2.10.1"
+version = "2.13.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e7/46/bd74733ff231675599650d3e47f361794b22ef3e3770998dda30d3b63726/pyjwt-2.10.1.tar.gz", hash = "sha256:3cc5772eb20009233caf06e9d8a0577824723b44e6648ee0a2aedb6cf9381953", size = 87785, upload-time = "2024-11-28T03:43:29.933Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/3b/81/58d0ac84e1ef3a3843791d6954d94c0b33d526c75eeb1efbce9d0a4c4077/pyjwt-2.13.0.tar.gz", hash = "sha256:41571c89ca91598c79e8ef18a2d07367d4810fbbd6f637794879baf1b7703423", size = 107515, upload-time = "2026-05-21T19:54:36.618Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/61/ad/689f02752eeec26aed679477e80e632ef1b682313be70793d798c1d5fc8f/PyJWT-2.10.1-py3-none-any.whl", hash = "sha256:dcdd193e30abefd5debf142f9adfcdd2b58004e644f25406ffaebd50bd98dacb", size = 22997, upload-time = "2024-11-28T03:43:27.893Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/5e/ecf12fdb62546d64385c158514e9b2b671f7832108ef2ecd2020ce0af2d1/pyjwt-2.13.0-py3-none-any.whl", hash = "sha256:66adcc2aff09b3f1bbd95fc1e1577df8ac8723c978552fd43304c8a290ac5728", size = 31274, upload-time = "2026-05-21T19:54:35.362Z" },
 ]
 
 [[package]]
@@ -2149,11 +2200,11 @@ wheels = [
 
 [[package]]
 name = "python-dotenv"
-version = "1.1.1"
+version = "1.2.2"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/f6/b0/4bc07ccd3572a2f9df7e6782f52b0c6c90dcbb803ac4a167702d7d0dfe1e/python_dotenv-1.1.1.tar.gz", hash = "sha256:a8a6399716257f45be6a007360200409fce5cda2661e3dec71d23dc15f6189ab", size = 41978, upload-time = "2025-06-24T04:21:07.341Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/82/ed/0301aeeac3e5353ef3d94b6ec08bbcabd04a72018415dcb29e588514bba8/python_dotenv-1.2.2.tar.gz", hash = "sha256:2c371a91fbd7ba082c2c1dc1f8bf89ca22564a087c2c287cd9b662adde799cf3", size = 50135, upload-time = "2026-03-01T16:00:26.196Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/5f/ed/539768cf28c661b5b068d66d96a2f155c4971a5d55684a514c1a0e0dec2f/python_dotenv-1.1.1-py3-none-any.whl", hash = "sha256:31f23644fe2602f88ff55e1f5c79ba497e01224ee7737937930c448e4d0e24dc", size = 20556, upload-time = "2025-06-24T04:21:06.073Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/d7/1959b9648791274998a9c3526f6d0ec8fd2233e4d4acce81bbae76b44b2a/python_dotenv-1.2.2-py3-none-any.whl", hash = "sha256:1d8214789a24de455a8b8bd8ae6fe3c6b69a5e3d64aa8a8e5d68e694bbcb285a", size = 22101, upload-time = "2026-03-01T16:00:25.09Z" },
 ]
 
 [[package]]
@@ -2167,11 +2218,11 @@ wheels = [
 
 [[package]]
 name = "python-multipart"
-version = "0.0.20"
+version = "0.0.32"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/f3/87/f44d7c9f274c7ee665a29b885ec97089ec5dc034c7f3fafa03da9e39a09e/python_multipart-0.0.20.tar.gz", hash = "sha256:8dd0cab45b8e23064ae09147625994d090fa46f5b0d1e13af944c331a7fa9d13", size = 37158, upload-time = "2024-12-16T19:45:46.972Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/5b/42/55c32bb9b12693c092ad250a0e82edb5b31ddeda6eb772de5f308b3804ad/python_multipart-0.0.32.tar.gz", hash = "sha256:be54b7f3fa167bb83e4fcd936b887b708f4e57fe75911c02aebf53efaf8d938e", size = 46881, upload-time = "2026-06-04T16:18:58.647Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/45/58/38b5afbc1a800eeea951b9285d3912613f2603bdf897a4ab0f4bd7f405fc/python_multipart-0.0.20-py3-none-any.whl", hash = "sha256:8a62d3a8335e06589fe01f2a3e178cdcc632f3fbe0d492ad9ee0ec35aab1f104", size = 24546, upload-time = "2024-12-16T19:45:44.423Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/04/e8135ebd1ad02c56ec633277529b2602ff99ff634be76cdba5744cf554fd/python_multipart-0.0.32-py3-none-any.whl", hash = "sha256:ff6d3f776f16878c894e52e107296ffc890e913c611b1a4ec6c44e2821fe2e23", size = 30042, upload-time = "2026-06-04T16:18:57.319Z" },
 ]
 
 [[package]]
@@ -2908,11 +2959,11 @@ wheels = [
 
 [[package]]
 name = "urllib3"
-version = "2.6.3"
+version = "2.7.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/c7/24/5f1b3bdffd70275f6661c76461e25f024d5a38a46f04aaca912426a2b1d3/urllib3-2.6.3.tar.gz", hash = "sha256:1b62b6884944a57dbe321509ab94fd4d3b307075e0c2eae991ac71ee15ad38ed", size = 435556, upload-time = "2026-01-07T16:24:43.925Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/53/0c/06f8b233b8fd13b9e5ee11424ef85419ba0d8ba0b3138bf360be2ff56953/urllib3-2.7.0.tar.gz", hash = "sha256:231e0ec3b63ceb14667c67be60f2f2c40a518cb38b03af60abc813da26505f4c", size = 433602, upload-time = "2026-05-07T16:13:18.596Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/39/08/aaaad47bc4e9dc8c725e68f9d04865dbcb2052843ff09c97b08904852d84/urllib3-2.6.3-py3-none-any.whl", hash = "sha256:bf272323e553dfb2e87d9bfd225ca7b0f467b919d7bbd355436d3fd37cb0acd4", size = 131584, upload-time = "2026-01-07T16:24:42.685Z" },
+    { url = "https://files.pythonhosted.org/packages/7f/3e/5db95bcf282c52709639744ca2a8b149baccf648e39c8cc87553df9eae0c/urllib3-2.7.0-py3-none-any.whl", hash = "sha256:9fb4c81ebbb1ce9531cce37674bbc6f1360472bc18ca9a553ede278ef7276897", size = 131087, upload-time = "2026-05-07T16:13:17.151Z" },
 ]
 
 [[package]]