Langfuse

Open-source LLM engineering platform: traces, evals, prompt management and metrics for teams that want LangSmith-class observability without locking into a closed vendor.

Product preview

Overview

Langfuse is the open-source answer to “we need traces and evals but we will not send production prompts to a closed SaaS.” It captures every LLM call, tool hop and retrieval, versions prompts, and runs datasets so quality is measured rather than guessed — then you self-host it next to the app or use their cloud.

Compared with LangSmith it is less tightly bound to LangChain and more MIT-licensed; compared with Helicone it goes beyond a logging proxy into prompt CMS and evals. The trade is that self-hosting is another service to run, and the cloud product is younger than Smith. For teams that treat observability as infrastructure they own, Langfuse is the default.

Pros / Cons

Pros

MIT-licensed traces you can self-host
Prompt CMS plus evals in one product
Framework-agnostic SDKs
Cloud option when you do not want to operate it

Cons

Self-hosting is another service to run
Cloud is younger than LangSmith
UI polish trails the closed suites

Who it's for

Tracing production LLM apps you cannot send off-site

Versioning prompts across a team

Running evals on golden datasets

Replacing a closed observability vendor

Bottom line

Open-source LLM observability you can actually own — the LangSmith alternative for teams that self-host.

Key Features

Open-source traces for LLM apps

Prompt versioning & playground

Datasets and evals

Self-host or Langfuse Cloud

SDK for Python, JS and more

Compare Langfuse with alternatives

Langfuse vs Hugging Face Langfuse vs Replicate Langfuse vs Ollama 4 products · AI Platforms

Frequently asked questions

What is Langfuse?

Langfuse is a AI Platforms tool developed by Langfuse. The company was founded in 2022 and is headquartered in Berlin, Germany. Open-source LLM engineering platform: traces, evals, prompt management and metrics for teams that want LangSmith-class observability without locking into a closed vendor.