Langfuse
Open-source LLM observability — tracing, evals, cost tracking, prompt management
📋 Overview
Langfuse is an open-source AI engineering platform that provides complete observability for LLM applications — tracing every call, evaluating outputs, tracking costs, and managing prompts. Unlike LangSmith (LangChain proprietary) or Helicone (observability-only), Langfuse combines tracing, evaluation, prompt management, and datasets in one platform. It integrates with OpenTelemetry, LangChain, OpenAI SDK, LiteLLM, and 20+ frameworks. Self-hosted is free; cloud adds team collaboration and advanced analytics.
✨ Key Features
- •31K+ GitHub stars — open-source Datadog for AI
- •Tracing: capture prompts, responses, latency, token usage
- •Evaluation: LLM-as-judge, heuristic, human evals
- •Prompt management: version and deploy prompts
- •Cost tracking: per-model, per-user analytics
- •Integrations: OpenTelemetry, LangChain, OpenAI, LiteLLM
- •Self-hosted free; cloud for teams
🎯 The Problem It Solves
Production LLM applications have no visibility into what models are doing — prompts, responses, costs, and quality are invisible. Langfuse provides Datadog-level observability for LLMs, with tracing, evaluation, and cost tracking in one open-source platform.
🔧 How It Works
Langfuse integrates via SDK or OpenTelemetry — wrap your LLM calls with Langfuse tracers to automatically capture prompts, responses, latency, token usage, and costs. The evaluation framework runs automated tests (LLM-as-judge, heuristic, human) against your traces. The prompt management system versions and deploys prompts independently of code. Datasets collect real-world inputs for testing. The dashboard provides trace exploration, cost analytics, and quality metrics.
🚀 Installation & Quick Start
Installation
See websiteQuick Start
- See documentation
✅ Pros
- •Complete LLM observability platform
- •Tracing captures every LLM call automatically
- •Built-in evaluation framework
- •Prompt management with versioning
- •Cost tracking and analytics
- •Self-hosted is free
- •Active YC-backed development
❌ Cons
- •Self-hosting requires PostgreSQL + ClickHouse
- •Free cloud tier tight limits (50K traces/mo)
- •Complex evaluations need LLM-as-judge setup
- •Documentation lags behind features
- •Some integrations are community-maintained
💬 Practitioner Verdict
“Langfuse is the best open-source LLM observability platform — the tracing is comprehensive, the evaluation framework is genuinely useful, and the prompt management is a differentiator. The trade-off: self-hosting requires PostgreSQL and ClickHouse, the free cloud tier has tight limits, and complex evaluations require LLM-as-judge setup. For teams shipping production LLMs, Langfuse is the default observability tool.”
Self-Hosted (Free)
Open source, MIT/Apache licensed. Run it yourself.
⭐ Star & Clone on GitHubFree forever. Your infrastructure, your data.
Langfuse Cloud Free
50K traces/mo
- 50K traces
- Basic features
- Community support
Langfuse Cloud Team
Unlimited traces, team features
- Unlimited traces
- Team collaboration
- Advanced evals
Deployment Options
Ways to run Langfuse in production
📊 Specifications
- Language
- TypeScript
- License
- MIT
- Platform
- Linux, macOS, Windows
- Supported Models
- REST API, CLI
💰 Pricing Reality
Self-hosted is 100% free MIT license. Cloud: Free tier (50K traces/mo), Team 9/mo (unlimited traces, team features), Enterprise custom (SSO, VPC, dedicated). No per-trace fees.