Langfuse is an open-source platform for observability and evaluation of AI agents and LLM applications. It records hierarchical traces of every model call, tool invocation and retrieval step, and lets teams filter by user, session, cost, latency or custom metadata. Around tracing it adds prompt management, a playground, datasets, experiments, human annotation and LLM-as-a-judge evaluators. Langfuse Cloud has a free Hobby tier with 50,000 units per month, Core at $29 per month, Pro at $199 per month and Enterprise at $2,499 per month, and it can also be self-hosted under an MIT license.
The product is built for AI engineers, ML platform teams and product teams who ship agents and need to understand what happens in production. Startups can begin on the free Hobby plan, growing teams move to Core or Pro for longer data retention, and large organizations choose Enterprise for audit logs, SCIM and uptime guarantees. Teams with strict data residency needs can self-host with Docker Compose, Kubernetes or Terraform on AWS, GCP and Azure. Domain experts who are not developers can join through annotation queues and the prompt interface without touching application code.
A developer debugging a failing agent opens a trace to see exactly which tool call or retrieval step went wrong. A product team separates prompts from code so they can deploy and roll back changes with one click. QA engineers build regression datasets from production traces and run experiments to compare models side by side before release. Finance-minded leads monitor token spend, cost and latency on dashboards with alerts. Because Langfuse is OpenTelemetry native with native Python and TypeScript SDKs and more than 100 integrations, it fits stacks built on LangChain, LiteLLM, Vercel AI SDK, CrewAI and many model providers.
