When an AI application fails in production, without observability you are in the dark. Langfuse is the open-source platform that lights it up: it traces every model and agent call, with cost, latency and tokens, plus evaluation and prompt management. It is the open alternative to LangSmith.
What is Langfuse?
Langfuse offers nested tracing of LLM and agent calls, prompt management with versioning, a playground and dataset/experiment tooling for evaluation. It has Python and JS SDKs and OpenTelemetry integration, and is self-hostable.
Key features
- Tracing of LLM and agent calls with cost, latency and tokens
- Prompt management with versioning and a playground
- Evaluation via LLM-as-judge, feedback and custom metrics
- Python/JS SDKs and OpenTelemetry integration, self-hostable
- Extremely active, with hundreds of releases
How Reche uses it
Observability from the first sprint is a RecheOS principle. Reche instruments the products it builds with tools like Langfuse, so AI in production is monitorable and improvable, not a black box.