Langfuse

Langfuse

By Langfuse

Open-source tracing, evaluation, and analytics for LLM and agent applications

v3.224.02026-07-22TypeScriptMIT
31.7k Open source Free tier Paid

Doc version 1.0 · Updated

Related concepts

Overview

Agents fail in ways that are hard to see from logs alone.
Langfuse records each run as a trace of nested spans (model calls, tool calls, retrievals) so you can inspect latency, cost, and the exact prompts/outputs at every step — then attach scores and evals on top.

flowchart LR
  run[agent run] --> trace[trace]
  trace --> gen[generation span]
  trace --> tool[tool span]
  trace --> ui[(Langfuse UI)]

The Code samples tab shows the TypeScript SDK and the Python decorator — pick from the selector to compare.

When to use it

Add Langfuse early when you need to debug multi-step agent runs, compare prompt versions, or track cost and quality in production.