AgentDbg – Local debugging for AI agent runs

Summary: AgentDbg records AI agent executions locally as JSONL files and provides a timeline UI to inspect LLM calls, tool calls, state changes, errors, and loop warnings. It supports any Python environment with optional LangChain or LangGraph integration and exports runs to a single JSON file without cloud dependencies.

What it does

AgentDbg captures detailed traces of agent activity including calls, errors, and state, displaying them in a local web interface. It features loop warnings, default redaction, truncation limits, and optional callback handlers for LangChain and LangGraph.

Who it's for

Developers debugging AI agents in Python who need local, detailed tracing without relying on cloud services or accounts.

Why it matters

It addresses the difficulty of diagnosing loops, stalls, and incorrect tool calls by providing a local debugger focused on agent behavior rather than dashboards.