Langfuse

Langfuse is the open-source platform for seeing what your LLM application actually did. Its tracing captures every model call, retrieval step, and agent action across multi-turn conversations and complex agent workflows, while prompt management adds versioning, deployment, and a built-in playground, and the evaluation layer supports LLM-as-judge, code evaluators, user feedback, and manual review. It is OpenTelemetry-native to reduce vendor lock-in and integrates with the OpenAI SDK, LangChain, LlamaIndex, LiteLLM, and 50+ other libraries via Python and JavaScript SDKs. You can self-host the full stack or use their cloud. Once an agent system grows past a prototype, this is the observability layer that makes debugging it tractable.