Hi everyone!
I built Agent-Devtools — a lightweight Python project to simplify debugging and tracking AI agents without the bloat.
Causal Debugging: Track memory influence, retrieval winners, injected context, and tool calls.
Behavior Diffing: Compare a "good" vs. "bad" run to pinpoint the exact root cause of divergence.
Deterministic Replay: Re-play recorded events offline to verify memory & retrieval consistency.
Context Provenance: Inspect the exact final LLM prompt with tags for every context source.
100% Local-First: Stores data in local SQLite with an auto-opening FastAPI dashboard (zero API keys needed).
Integrations: Native callbacks for LangChain, Groq, and custom Python agent loops.
GitHub Repo:https://github.com/Jacopos311/Agent-Devtools
Top comments (0)