AI agent observability
August 2026
Agent Tracer
A developer platform that makes AI-agent executions understandable, from the first model call to the last tool response.
- Live timelines for every LLM and tool call
- Model-call replay and trace comparison
Built with: Python · FastAPI · PostgreSQL · React · TypeScript · OpenTelemetry
Technical details for Agent Tracer
- Built an observability platform for AI agents: a Python SDK traces LLM and tool calls as nested spans, and an OTLP/HTTP endpoint ingests OpenTelemetry/OpenInference traces into a FastAPI and PostgreSQL backend.
- Developed a React/TypeScript dashboard with waterfall span timelines, per-step input/output/error inspection, and latency, token, and cost tracking, updated live via Server-Sent Events.
- Implemented side-effect-free replay of individual model calls against OpenAI/Anthropic, plus trace comparison that aligns span trees to surface status flips, drift, and latency/cost deltas.
- Added hashed API keys, Docker Compose, and GitHub Actions CI running 74 tests on SQLite and PostgreSQL.