Langfuse is the open-source LLM engineering platform for tracing, evaluating and improving AI agents, connecting observability, prompt management, datasets and experiments in one loop.
LLM applications fail in ways traditional monitoring cannot see: every step carries rich, verbose input and output that legacy observability tools were not built to store or query. Langfuse instruments your application — via OpenTelemetry or one of 100+ integrations with frameworks and model providers — and records full traces of every call, so you can inspect exactly what an agent saw, decided and returned, along with its cost and latency.
On top of tracing, Langfuse closes the improvement loop. Teams manage and version prompts, build datasets from real production traces, run offline experiments and online evaluations (including LLM-as-a-judge and human annotation), and feed scores back into development. A built-in assistant, Halo, can investigate production issues and take approved actions, and CLI and MCP access let coding agents work with Langfuse data directly. Because the platform is open source, you can self-host it at scale and keep complete control of your data, or use the managed cloud.
The free tier includes 50,000 observations per month with no credit card required, and the open-source version can be self-hosted for free. Paid cloud plans start at $29/month for higher volumes and team features, scaling to enterprise tiers with SSO, dedicated support and compliance requirements. Pricing scales primarily with the number of observations ingested.
Teams shipping LLM features and agents who need to debug behavior and measure quality systematically. One-off scripts or prototypes that never reach real users rarely need this level of observability.