Langfuse

Langfuse helps teams debug LLM applications and AI agents, then test changes against production examples. It brings traces, evaluations and prompt revisions into one place.

• Traces and monitoring: Inspect model calls, tool actions and retrieval steps, filter by user, session, cost or latency, and track problems with dashboards and automated alerts.
• Evaluations and experiments: Score outputs with model judges, code checks or human review, build datasets from reviewed traces, and compare results against test cases.
• Prompt management: Keep prompts separate from application code, collaborate on versions and roll back deployments.
• Model comparison: Test prompts with production inputs in the playground and compare model responses side by side.
• Coding agent access: A CLI and MCP server let coding agents query traces, manage prompts and work with evaluations.

Langfuse is open source under MIT and free to self-host on your own servers with Docker Compose or Kubernetes. Langfuse Cloud provides a hosted alternative. It supports OpenTelemetry and native Python and TypeScript SDKs, with integrations including Ollama, vLLM, OpenWebUI, LangChain and LiteLLM. REST APIs and data exports provide access outside the interface.