Langfuse

Open-source LLM observability: traces, scores, prompt mgmt, self-hostable

MLOps & LLMOps · Intermediate · ~5 min

In plain English

Open-source tracing and evaluation for LLM apps — the same job as a hosted observability tool, on infrastructure you control.

Why it's worth your time

When traces contain regulated data, self-hosting isn't a preference, it's the requirement.

If you remember three things

Traces, prompt management, evals and cost tracking in one place
Self-hostable, which is often the deciding factor
Prompt versioning outside code lets non-engineers iterate safely

Overview

Langfuse is an open-source platform for engineering LLM applications — the leading open, self-hostable alternative to LangSmith. An SDK or framework callback captures traces and spans with cost and latency, you attach scores from evals or user feedback, and you manage versioned prompts and datasets. Analytics roll everything into trends over time, and because it is Apache-2.0 you can self-host it and keep all trace data on your own infrastructure.

In an interview

Langfuse is an open-source LLM engineering platform: tracing, evaluation scores, prompt management, and analytics. You drop in the SDK — an @observe decorator or a framework callback — and every call is captured as a trace of nested spans with token cost and latency. You attach scores from model-based evals or user feedback, manage versioned prompts and datasets, and watch cost and quality trends over time. Its differentiator is being Apache-2.0 and self-hostable, so it's the open alternative to LangSmith with your data on your own infra.

Production defaults

Self-host: when trace content is sensitive. That is the main reason to choose it
Prompts: versioned with the eval score attached. A prompt without a score is a guess
Sampling: all failures, sampled successes
Portability: keep prompts mirrored in your repo so the tool stays replaceable

What breaks

Tracing added latency — Synchronous export. Batch and send asynchronously — never on the request path.
Prompt changed in the UI and production shifted — Prompts in a console are unreviewed deploys. Gate them on evals like any other change.

Watch it explained

Langfuse — The Open Source LLM Observability Platform Explained — Prism Labs, 7:18