langfuse/langfuse

Langfuse is an open-source LLM engineering platform that handles the unglamorous but critical layer between your AI feature working in a demo and working in production. It covers observability, evals, prompt versioning, datasets, and a playground — all in one self-hostable stack. What pushes it toward a pick: the breadth is real, not cosmetic. OpenTelemetry support means you are not locked into a proprietary tracing SDK. Native integrations with LangChain, OpenAI SDK, and LiteLLM means you drop it in without rewriting your call sites. The prompt management piece alone is worth the install if you have ever hotfixed a production prompt by redeploying your entire backend. The cloud version exists if you do not want to run it yourself. Honest reservation: if your LLM surface is one endpoint that barely gets traffic, this is overkill. Once you hit three models and a handful of prompt variants, it earns its place fast. -> Best for: AI engineer or SaaS team of 2-5 shipping LLM features to production