Langfuse is an open-source LLM engineering platform for tracing, prompt management, evaluation, and dataset curation. It works with any framework (LangChain, LlamaIndex, OpenAI SDK, custom) via lightweight SDKs.
Core capabilities
- Tracing — nested spans for every LLM/tool call, with token and cost tracking.
- Prompts — versioned prompt library with A/B testing.
- Evals — LLM-as-a-judge, custom Python evals, human annotation queues.
- Datasets — curate golden sets from production traces for regression testing.
When to reach for it
- Any agent going to production — you need traces to debug and evals to prevent regressions.
- Self-hosted deployments in regulated industries.
Learn more
- Site: <https://langfuse.com>
- Docs: <https://langfuse.com/docs>
- Repo: <https://github.com/langfuse/langfuse>