etrace: open source minimal tracing lib

etrace is an open-source AI agent tracing library that auto-instruments LLM calls and tracks costs across more than 1,700 models, built for developers debugging and observing AI workflows. Its design centers on a single primitive, trace, with everything else handled automatically: when provider packages like OpenAI or Anthropic are installed, LLM calls are traced without extra wiring. It classifies operations using 16 semantic span kinds, including workflow, agent, step, llm, tool, retrieval, embedding, guardrail, eval, and handoff, so traces map to how agents actually run. Costs are auto-calculated from a bundled pricing catalog covering 1,700+ models across 100+ providers, and the cost auto-populates when you set usage on a span. etrace Studio is a local companion that lets you visualize, inspect, and debug traces, receiving data over OTLP at a local endpoint. Installation is a single package: pip install etrace-py for Python or npm install @evaris/etrace for TypeScript, with an optional OpenTelemetry export extra. The library is MIT licensed and free. It is best for engineers who want lightweight, local-first observability for LLM agents without a hosted backend.

Tracing, debugging, and cost-tracking LLM calls in AI agent workflows