# Langfuse
Langfuse is an open-source LLM observability platform. Traces, evals, prompt management, datasets, scores — the same surface area as [[LangSmith]], but you can run the whole thing on your own server. SDKs are MIT; the server is MIT for the core and commercial for some enterprise extras.
This is the one I reach for when I want LLM-app observability without sending prompts and traces to a third party.
## What it gives you
- **Tracing** — nested traces of LLM calls, tools, retrievals, and arbitrary spans. Framework-agnostic, with first-class integrations for LangChain, LlamaIndex, OpenAI SDK, Anthropic SDK, and OpenTelemetry.
- **Prompt management** — versioned prompts with environments (production, staging) and pull-by-label from code. Built for the "edit prompt without redeploy" workflow.
- **Datasets + evals** — pull traces into datasets, run experiments, compare runs side by side. LLM-as-judge, code-based scorers, or human feedback.
- **Sessions + users** — group traces by session ID or user ID for conversation-level analysis.
## When I'd reach for it
- You want LLM observability and you want to self-host.
- You need a real prompt management layer (versions, labels, environments) and don't want to roll your own.
- You're comparing LangSmith but the hosted-only default is a dealbreaker.
## Trade-offs
- **Self-hosting has a cost.** Postgres, ClickHouse, Redis, queue workers — it's a real deployment, not a single binary. Their cloud removes that pain (and pays the team).
- **The UI is good but not flashy.** Optimized for engineers debugging traces, not for execs reading dashboards.
- **Some advanced features are cloud-only** (SSO, advanced RBAC). Read the matrix before you commit to self-host.
## Alternatives
- [[LangSmith]] — closed source, polished, tied to LangChain's brand.
- [[Helicone]] — open source too, but proxy-first. Easier 1-line install; less prompt-management depth.
- [[MLflow]] — wider scope (classic ML + LLM); LLM features are newer and less specialized.
- [[Edgee]] — observability via gateway pattern, optimized for cost/routing more than evals.
## References
- Website: https://langfuse.com/
- GitHub: https://github.com/langfuse/langfuse
- Docs: https://langfuse.com/docs
## Related
- [[LangSmith]]
- [[Helicone]]
- [[LLM Monitoring]]
- [[AI Observability]]
- [[MLflow]]
- [[Edgee]]
- [[LangChain]]
- [[LangGraph]]