Skip to content
ML2
Multi-Step Agent SetupsToolobservabilityevalsopen source

Langfuse: tracing and evals for agents

Added

Our verdict

Once an agent runs more than a few steps, you need traces to debug it. Langfuse gives you that, plus evals to catch regressions when you change prompts or models. Self-host it or use the cloud; add it before you need it, not after.

Who it's for
Builders running agents in production or on a schedule
Leverage
Debug and improve agents from real traces
Effort
Medium
Date added

What it does

Langfuse is an open-source platform for agent evals and observability: trace, evaluate and improve LLM applications in one place. Tracing is built on OpenTelemetry with native Python and JS SDKs, and evaluation supports LLM-as-a-judge, user feedback, manual annotation and datasets for experiments.

How builders use it

  • Instrument your agent so every run produces a trace of prompts, tool calls and costs.
  • Turn bad runs into a dataset and score new prompt versions against it.
  • Watch cost and latency per step to find the expensive ones.

What to expect

You stop guessing why an agent failed and can prove a change made it better.

Related discoveries

Want more like this?

Practical discoveries, without the noise.

Get the newsletterMore in Multi-Step Agent Setups