Langfuse

Founded by Clemens Rawert & Marc Klingen & Max Deichmann in 2023

Trace, evaluate, and monitor complex LLM applications and autonomous AI agents in production

Pricing model

Free Tier, Paid

Rating

★ People love it

Time to value

Quick Setup (< 1 hour)

You can use Langfuse to trace, evaluate, and monitor your language model applications and AI agents from early development through production. It captures every model call, tool invocation, and retrieval step to help you debug errors, analyze latency, and track token costs. You can manage and version prompts outside your codebase, run automated evaluations with model-based or custom scoring functions, and review production traces with human annotation workflows. It connects to your existing frameworks through open standards and provides test playgrounds to compare model outputs.

What Langfuse does

Trace every function and model call in an agentic workflowCalculate exact token spending per user session or featureDeploy updated system prompts across active production environmentsRun automated regression tests on prompt changesFlag and inspect low-scoring or failed AI generationsExport trace telemetry to external data warehouses or S3 bucketsCapture detailed hierarchical traces for every model call, tool execution, and retrieval stepManage, version, and deploy prompts without modifying application codeTrack token usage, API expenses, and latency across different model providersEvaluate outputs using model-as-a-judge evaluators, custom code rules, or human reviewsTest and compare prompts side by side in an interactive playgroundExport production traces to external blob storage or query them directly via APISelf-host using Docker or Kubernetes with an open-source MIT license