Langfuse is the most popular open-source LLM observability platform, providing tracing, evaluation, and monitoring for AI applications. With over 21,000 GitHub stars and acquired by ClickHouse for its analytical capabilities, it has become a critical part of the LLM engineering stack.
The tracing system captures detailed information about every LLM call including inputs, outputs, latency, token usage, and costs. Complex agent workflows with multiple LLM calls are visualized as nested traces, making it easy to debug and optimize multi-step applications.
Prompt management with versioning allows teams to iterate on prompts, track changes, and deploy specific versions to production. Dataset-based evaluation enables systematic testing with custom metrics and LLM-as-judge evaluators. User feedback collection creates ground truth for continuous improvement.
Langfuse is framework-agnostic with native integrations for LangChain, LlamaIndex, OpenAI SDK, Vercel AI SDK, and more. Cost tracking aggregates LLM spending across models and provides breakdowns by user, feature, or time period.
Both self-hosted and managed cloud deployment are available. The open-source version can be deployed via Docker with full feature parity. The managed cloud offers additional convenience with automatic updates and scaling.