Langfuse

What it is

Open-source tracing, prompts, and evals.

Langfuse is the open-source observability leader, with 28,000+ GitHub stars and an MIT licence. It gives you full tracing so you can see every request, prompt management and versioning with a built-in playground, and evaluation via LLM-as-judge, user feedback, or custom metrics, with multi-turn support. You can self-host it freely, so nothing about your traces or prompts leaves infrastructure you control.

The one job

See, version, and evaluate, on your own infrastructure.

See every request as a trace, version your prompts over time, and run evals against them, all in one place you own. No vendor holds your data hostage.

Reach for it when

• You want open-source, self-hosted observability with no vendor lock-in.
• You need prompt versioning and a playground alongside your traces.
• You want tracing and evals in one place rather than two tools.
• You care about owning your data: traces and prompts stay on your infrastructure.

Skip it when

• You are deep in LangChain and want the richest native integration: LangSmith wins there.
• You want managed-only with zero ops and no interest in self-hosting.
• You need a large prebuilt metric library out of the box: Phoenix ships more graders.

A minimal look

Wrap the call in a trace.

import { Langfuse } from "langfuse";
import OpenAI from "openai";

const lf = new Langfuse();          // points at your self-hosted host
const openai = new OpenAI();
const trace = lf.trace({ name: "answer" });

const gen = trace.generation({ name: "chat", model: "gpt-4o" });
const res = await openai.chat.completions.create({
  model: "gpt-4o", messages,
});
gen.end({ output: res.choices[0].message });  // span is now in Langfuse

The principle it teaches

Where it fits your labs

This is a Week 5 tool. When you instrument your agent for tracing and wire up graders in Lab 5, Langfuse is where the traces land and where your evals run.