DIDACTYL

Status Dashboard

Agent health, runtime info, and API connectivity.

API Protocol

API Host

API Port

API: Unknown

API Debug Log

Context Inspector

Inspect current LLM context parts with token estimates and content.

No context loaded.

Simple Prompt Crafting

Run a basic system + user prompt without tools.

System Prompt

User Message

Model Override (optional)

model_used: -

input_tokens_estimate: -

output_tokens_estimate: -

Full Prompt with Tools

Edit the full messages array and run with tool execution enabled.

Tool execution is real when calling /api/prompt/run; agent tool calls can perform real side effects.

Messages JSON Array

Max Turns

Model Override (optional)

I understand this can execute real tools.

model_used: -

total_input_tokens_estimate: -

total_output_tokens_estimate: -

Final Response

Turn Trace / Tool Calls

A/B Prompt Comparison

Run two prompt variants side-by-side and compare outputs.

Comparison also executes tools for each variant when model behavior chooses tool calls.

Variant A Messages JSON

Max Turns A

Model A (optional)

Variant B Messages JSON

Max Turns B

Model B (optional)

Variant A stats: -

Variant B stats: -

Variant A Final Response

Variant B Final Response

Difference view is text-first; copy results into external diff tooling if you need strict semantic diffs.