Skip to content
OpenSourceSoftware.ai
Langfuse logo

Langfuse

langfuse · Observability

GitHub stars

36,610
—Awaiting first sync

Awaiting first sync

No measured star history yet

What is Langfuse?

Langfuse gives teams building with large language models the same visibility that error tracking gives web developers. Every model call can be traced, evaluated and compared, so improving an AI feature becomes measurable instead of guesswork.

Langfuse is an open-source LLM engineering platform for developing, monitoring, evaluating and debugging AI applications. It records traces of model calls, retrieval and agent actions, and adds prompt management, evaluations, datasets and a playground. It can be self-hosted with Docker or used in Langfuse Cloud.

Core capabilities

  • Trace LLM calls, retrieval, embeddings and agent actions across your app.
  • Version and manage prompts centrally, and test them in a playground.
  • Evaluate quality with LLM-as-a-judge, code evaluators, user feedback and datasets.

Observability for LLM applications

Applications send traces of LLM calls and related logic such as retrieval, embeddings and agent actions. Teams can inspect complex runs, follow user sessions and debug failures, with costs and latencies recorded alongside each step.

  • Tracing
  • LLM observability
  • Debugging

Prompts, evaluations and datasets

Prompt management versions prompts centrally with caching so changes ship without latency costs. Evaluations support LLM-as-a-judge, code evaluators, user feedback and manual labelling, and datasets hold test sets for experiments before deployment.

  • Prompt management
  • Evaluations
  • Datasets

Self-host in minutes

Langfuse runs locally with Docker Compose or on your own infrastructure, and Langfuse Cloud offers a managed version with a free tier. An OpenAPI specification and typed SDKs support custom LLMOps workflows.

  • Self-hosted
  • Docker
  • API

Where it fits

Use cases

  1. 01

    See what an AI feature is doing in production

    Instrument the app with an SDK and follow each request through model calls, retrieval and tool use to find slow or wrong answers.

  2. 02

    Change prompts without redeploying

    Keep prompts in Langfuse, version them, and roll out an improved version after testing it in the playground.

  3. 03

    Check quality before a release

    Run a dataset of test cases through a new prompt or model and score the results with LLM-as-a-judge or code evaluators.