Langfuse
Open-source LLM observability platform for debugging and improving AI applications
Visit Product
213 upvotes
2,238 views
About
Langfuse is an open-source LLM engineering platform designed to help teams build, debug, and improve LLM-powered applications. It provides comprehensive observability into every aspect of LLM application behavior — from prompt inputs and model outputs to latency, cost, token usage, and user feedback — giving developers the visibility they need to ship reliable AI products.
The platform allows developers to trace entire LLM application execution flows, including chains, agents, and retrieval steps, making it possible to identify exactly where errors occur, why responses are poor quality, or what's causing slow performance. Langfuse's evaluation framework lets teams systematically assess LLM output quality using automated scorers, human annotation, and custom metrics.
Beyond debugging, Langfuse enables prompt management — versioning, A/B testing, and deploying prompts without code deployments — and integrates with all major LLM frameworks including LangChain, LlamaIndex, OpenAI SDK, and others through native integrations or the OpenTelemetry standard.
The platform allows developers to trace entire LLM application execution flows, including chains, agents, and retrieval steps, making it possible to identify exactly where errors occur, why responses are poor quality, or what's causing slow performance. Langfuse's evaluation framework lets teams systematically assess LLM output quality using automated scorers, human annotation, and custom metrics.
Beyond debugging, Langfuse enables prompt management — versioning, A/B testing, and deploying prompts without code deployments — and integrates with all major LLM frameworks including LangChain, LlamaIndex, OpenAI SDK, and others through native integrations or the OpenTelemetry standard.
Product Features
- Distributed tracing for LLM chains, agents, and RAG pipelines
- Cost and token usage tracking across all LLM providers
- Prompt management with versioning and A/B testing
- Evaluation framework with automated and human-in-the-loop scoring
- Dataset management for regression testing
- User session tracking for conversational applications
- LangChain, LlamaIndex, OpenAI SDK native integrations
- OpenTelemetry-compatible for universal framework support
- Self-hostable with Docker Compose or Kubernetes
- Playground for prompt iteration and testing
- Cost and token usage tracking across all LLM providers
- Prompt management with versioning and A/B testing
- Evaluation framework with automated and human-in-the-loop scoring
- Dataset management for regression testing
- User session tracking for conversational applications
- LangChain, LlamaIndex, OpenAI SDK native integrations
- OpenTelemetry-compatible for universal framework support
- Self-hostable with Docker Compose or Kubernetes
- Playground for prompt iteration and testing
About the Publisher
Langfuse was founded in 2023 by Maximilian Deichmann and Clemens Rawert, two entrepreneurs from Berlin who previously built B2B SaaS products. The company is Y Combinator W23 batch and is backed by Lightspeed Venture Partners and General Catalyst. Langfuse has grown rapidly alongside the LLM application development ecosystem, with tens of thousands of developers using the platform to monitor production AI applications.