Keywords
LLM, Observability, Agent Tracing, AI, AgentOps

Technologies
Typescript, FastAPI, PostgreSQL, Python, React

TraceBrain is an open-source platform for monitoring, debugging, and improving AI agent systems. At its foundation is a centralized TraceStore which stores and manages all trace information, capturing agent execution traces at runtime and visualizing them as interactive, hierarchical trees, making it easy to see exactly what an agent did, which tools it called, and where it went wrong.

TraceBrain features an AI Librarian which allows users to query the TraceStore using natural language, and AI-generated trace assessments of agent behaviour through a human-in-the-loop labelling interface. Agents can escalate uncertainty at runtime, triggering alerts that surface directly in the dashboard for review.

Beyond observability, TraceBrain allows operators to retrieve past execution traces as references for trace improvement, and automatically generates targeted training tasks from recurring trace failures, turning past execution failures into actionable improvement material. Together, these capabilities form a closed loop where agent execution can be converted into structured learning that drives continuous agent evolution.