Langfuse

Langfuse is the open-source LLM engineering platform for observability, tracing, evaluation, and prompt management. It traces every LLM call, tool call, and agent step; runs evaluations with LLM-as-judge or datasets; manages prompt versions; and gives you dashboards, cost tracking, and alerts. It's the most popular self-hostable LLM observability tool, trusted by Shopify, MoonPay, and thousands of teams.

Key Features

• Full Tracing: Every LLM call, tool call, and agent step traced end-to-end — including LangGraph, CrewAI, and OpenAI Agents SDK.
• Evaluations: LLM-as-judge, manual scoring, and dataset-based evals in one place.
• Prompt Management: Versioning, registry, and rollout of prompts.
• Metrics & Dashboards: Latency, cost, token usage, and quality metrics out of the box.
• Cost Tracking: Per-model, per-session, and per-user cost breakdowns.
• Sessions & Users: Track sessions and user interactions for product analytics.
• Data Redaction: Automatic redaction of sensitive data in traces.
• 40+ Integrations: LangChain, LangGraph, OpenAI, Anthropic, LlamaIndex, CrewAI, and more; Python and TypeScript SDKs.
• Self-Host or Cloud: Docker self-hosting in one command, or managed cloud in EU/US regions.

Use Cases

Who Should Use This Tool?

• LLM Product Teams: Teams shipping LLM apps and agents who need to debug and improve them.
• Privacy-Conscious Companies: Organizations that must self-host observability for compliance.
• Prompt Engineers: Teams iterating on prompts and evals at scale.

Problems It Solves

1. Black-box LLM behavior: Full traces make every call and agent step inspectable.
2. Evaluation at scale: LLM-as-judge and datasets make quality measurable.
3. Prompt chaos: Versioned prompt management replaces copy-paste workflows.

Pricing

Advantages & Unique Selling Points

Compared to Competitors:

1. Open source and self-hostable: Full-featured observability without vendor lock-in.
2. Breadth of integrations: 40+ frameworks and SDKs, from LangChain to CrewAI to OpenAI.
3. All-in-one: Tracing, evals, and prompt management in a single platform.

What Makes It Stand Out:

• 15,000+ GitHub stars and the largest self-hosted LLM observability community.
• Built by a Berlin-based team (founded 2023) with enterprise adoption.
• From single-command Docker deploy to managed cloud.

Getting Started

Quick Start Guide

1. Deploy: Run the Docker image, or sign up for Langfuse Cloud.
2. Get keys: Create project API keys.
3. Integrate: Add the Langfuse SDK to your LLM calls (one line for OpenAI/Anthropic).
4. Observe: View traces, run evals, and manage prompts in the dashboard.

Integration

Integrates with:

• LangChain, LangGraph, LlamaIndex, CrewAI, OpenAI Agents SDK
• OpenAI, Anthropic, and 40+ other frameworks
• Python and TypeScript SDKs

Frequently Asked Questions

Is Langfuse free?

The core is open source and self-hostable for free; Langfuse Cloud has a free tier plus paid plans.

Can I self-host it?

Yes — a single Docker command deploys the full platform on your infrastructure.

Does it support agent frameworks?

Yes — it traces LangGraph, CrewAI, OpenAI Agents SDK, and many more.

Alternatives

If Langfuse isn't the right fit, consider these alternatives:

• LangSmith: LangChain's platform with deeper LangGraph deployment features.
• Helicone: LLM observability with a gateway for caching and rate limiting.
• AgentOps: Agent-focused session recording and replay.

Tips & Best Practices

1. Trace everything from day one: Add observability before you ship, not after.
2. Use LLM-as-judge evals: Automate quality scoring on real production traces.
3. Version your prompts: Use the prompt registry to roll back regressions.

Conclusion

Langfuse is the open-source LLM engineering platform that makes agents observable, evaluated, and prompt-managed — self-hostable or in the cloud. If your agents run in production and you need full visibility, Langfuse is the community standard.