Langfuse

What it's good for

• 1

  Self-hosted LLM observability where trace data has to stay in-house

• 2

  Tracing prompts, tool calls, retries, latencies, and token spend across a production app

• 3

  Running evals against a golden dataset to catch regressions before they ship

• 4

  Prompt management with versioning, so copy changes do not require a redeploy

• 5

  Collecting user feedback and scores, then feeding them back into eval sets

Pricing

• Hobby

  50k observations/mo, unlimited seats

  Free

• Core

  Higher limits and longer retention

  $29/mo

• Pro

  For growing teams

  $199/mo

• Enterprise

  SSO, support, compliance

  $2,499/mo

• Self-hosted

  Open source, MIT; enterprise self-host license $500/mo

  Free

How to use it

Wrap your LLM client with the Langfuse SDK (Python or JS) and every call is traced automatically. Tag traces with metadata so you can filter by user, feature, or environment. Manage prompts inside Langfuse and fetch them at runtime, so non-engineers can tune copy without a deploy. Build eval datasets from real traces, then run automated scoring (LLM-as-judge, exact match, or custom) against new prompt versions. For data-sensitive clients, self-host the open-source stack instead of using the cloud, and reach for the enterprise self-host license when you need SSO and support.

Pros & cons

Frequently asked questions

• Is Langfuse free?

  Yes. The self-hosted open-source version is free under an MIT license, and the cloud has a free Hobby tier with 50,000 observations a month and unlimited seats. Paid cloud runs $29/mo (Core), $199/mo (Pro), and $2,499/mo (Enterprise). The enterprise self-host license is $500/mo when you need SSO and support on your own deployment.

• Langfuse vs LangSmith: what is the difference?

  They do the same core jobs (tracing, evals, prompt management), but Langfuse is open source and self-hostable while LangSmith is a closed managed service from the LangChain team. If you want to own your trace data or keep it on your own infrastructure, Langfuse is the pick. If you are already deep in the LangChain paid ecosystem and want a turnkey hosted tool, LangSmith fits more naturally. Both work with any framework.

• Is Langfuse open source?

  Yes, under an MIT license. You can self-host the full platform for free and keep all observability data in your own environment. The cloud offering and the enterprise self-host license are paid layers on top of that same open-source core.

• What is Langfuse best for?

  Teams that want production LLM observability without shipping their trace data to a third party. It shines for self-hosted deployments, prompt management that non-engineers can edit, and eval workflows that catch regressions before release.

• Can I self-host Langfuse?

  Yes, and it is free to do so under the MIT license. You run it with Docker or on your own cluster and keep every trace in-house. If you need enterprise features like SSO on that self-hosted instance, there is a $500/mo license for it.

More

• Langfuse docs
• Langfuse pricing

Alternatives to Langfuse

Other tools we'd consider for the same job.

• LangSmith

  LLM observability, tracing, and evals from the LangChain team. The piece every serious AI deployment needs before it goes wrong in production

Other Data & ML Ops tools

• Qdrant

  Open-source vector database with a Rust engine. Self-host for data sovereignty, or run the managed cloud with a free tier

• Ollama

  Free, open-source runtime for local LLMs. Run Llama, DeepSeek, Mistral and more on your own machine with one command

• Braintrust

  Commercial eval and observability platform for LLM apps: evals, logging, and a prompt playground with unlimited users

• Pinecone

  Fully-managed vector database for production RAG. Serverless tier, low operational overhead — the default for client deployments