Langfuse

For teams that need to see what their LLM application did on each request, Langfuse records nested traces of model calls, retrieval steps, tool invocations and agent actions, then adds prompt versioning, test datasets, evaluations and a model playground. Instrumentation comes from Python and TypeScript SDKs, an @observe decorator, an OpenAI drop-in wrapper, and integrations for LangChain, LlamaIndex, Haystack, LiteLLM, the Vercel AI SDK, CrewAI and Mastra. Scores come from LLM judges, code checks, human labels or user feedback. A self-hosted install runs web and worker containers backed by Postgres, ClickHouse, Redis or Valkey, and S3-compatible storage; Docker Compose suits a laptop or single VM, and Helm on Kubernetes is the recommended production route, with Terraform templates for AWS, Azure and GCP. Code outside the ee folders is MIT and self-hosters get the core features; a few enterprise add-ons such as UI customization need a license key. The company joined ClickHouse in January 2026 and sells a managed cloud with a free tier. Suited to production LLM teams who want telemetry on their own servers.