Langfuse is an open-source LLM engineering platform for developing, monitoring, evaluating, and debugging AI applications. It provides observability for model calls and agent workflows, allowing teams to inspect traces, sessions, retrieval steps, embeddings, tool actions, and other application behavior when diagnosing quality, latency, or cost problems.

Beyond tracing, Langfuse includes centralized prompt management and versioning, datasets for repeatable testing, evaluation workflows, user feedback, manual labeling, LLM-as-a-judge and code-based evaluators, and a playground for experimenting with prompts and model configurations. It exposes APIs and typed SDKs and integrates with common AI stacks such as LangChain, LlamaIndex, LiteLLM, Vercel AI SDK, and other frameworks.

Langfuse can be used as a managed service or self-hosted with Docker, Kubernetes, Helm, Terraform, and major cloud platforms. It is aimed at teams moving from AI prototypes to production systems that need systematic observability and evaluation while retaining the option to operate the LLMOps stack on their own infrastructure.