"
Overview
Langfuse is an open-source observability and evaluation platform purpose-built for AI agents and LLM applications. It unifies tracing, prompt management, evaluation, experiments, and human feedback into a single continuous loop, so teams can launch, observe, and improve their AI systems with confidence. With v4 delivering up to 165× faster performance, Langfuse now handles 90B+ observations per month and is trusted by 21 of the Fortune 50 and over 100,000 engineers worldwide.
Developers love Langfuse because it works with any stack — supporting OTel instrumentation across Python, TypeScript, Go, Java, .NET, Ruby, PHP, and Swift, plus 100+ integrations with major agent frameworks (LangChain, CrewAI, LiteLLM, Pydantic AI, Google ADK) and model providers (OpenAI, Anthropic, Bedrock, Gemini, Mistral, Groq, vLLM). There is no framework lock-in, and the platform scales from prototype to production.
The integrated approach means teams can move from debugging a single trace to running structured experiments and shipping improved prompts — all in one place, with cost and latency dashboards keeping quality and spend under control.
Key Features
- Observability: Hierarchical traces capture every LLM call, tool invocation, and retrieval step, filterable by user, session, cost, latency, or custom metadata.
- Evaluation: Run LLM-as-a-judge, heuristic functions, or human review on production data or during experiments to measure quality continuously.
- Prompt Management: Separate prompts from code with one-click deployments and rollbacks, turning prompt improvement into a team sport.
- Playground & Experiments: Test prompts on real production inputs, compare models side-by-side, and run structured experiments with clear result comparisons.
- Human Annotation: Collaborative human-in-the-loop workflows to review traces and build golden datasets for training and evaluation.
- Cost & Latency Monitoring: Dashboards and automated alerts track spend, latency, and quality in real
