Langfuse Documentation — LLM Observability, Tracing and Evals
by Langfuse
Instrument your LLM app so every prompt, tool call and agent step is traced, scored and cost-attributed.
Overview
Langfuse is an open-source, self-hostable platform for debugging, analyzing and iterating on LLM applications, and its docs are organized around four pillars. Observability and tracing capture the full context of each request — prompts, responses, tool calls, retrieval steps, latency and token cost — including multi-turn sessions, per-user views and agent-graph visualization, so a failing agent run can be replayed step by step. Prompt management gives prompts version control, deployment labels and a playground, decoupling prompt changes from code deploys. Evaluation supports LLM-as-a-judge, custom code evaluators, human annotation and user feedback, run against datasets so you can compare versions rather than guess. Dashboards roll quality, cost and latency up into metrics you can watch over time. Instrumentation is via native Python and JavaScript SDKs plus OpenTelemetry, with 100+ integrations including the OpenAI SDK, LangChain, LlamaIndex and LiteLLM. The platform is free to self-host under an open-source license; a managed cloud offering with a free tier and paid plans is also available.
At a Glance
- Topic
- Frameworks
- Level
- Intermediate
- Format
- Documentation
- Cost
- Freemium
- Duration
- ~2-3 hours to instrument a first app
- Provider
- Langfuse
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓How to instrument an LLM app or agent so every span — prompt, tool call, retrieval, model response — is traced
- ✓How to debug a failing agent run by replaying its trace step by step with inputs, outputs and latency
- ✓How to attribute token cost and latency per user, session, feature or model
- ✓How to version, label and deploy prompts independently of application code
- ✓How to run LLM-as-a-judge and custom evaluators against datasets to catch quality regressions
- ✓How to instrument via OpenTelemetry or a first-party integration (OpenAI SDK, LangChain, LlamaIndex, LiteLLM)
Highlights
- •Fully open source and self-hostable — you can keep prompts and traces inside your own infrastructure
- •OpenTelemetry-native, so it fits an existing observability stack rather than replacing it
- •Tracing, prompt management and evals in one system instead of three disconnected tools
- •100+ framework and SDK integrations, including agent frameworks and LLM gateways
Who It’s For
Best For
- ✓AI engineers taking an LLM app or agent from prototype to production
- ✓Teams that need cost, latency and quality metrics per feature or per customer
- ✓Anyone who has to debug a nondeterministic multi-step agent failure after the fact
Prerequisites
- •Python or TypeScript/JavaScript
- •An LLM application you can instrument
- •Basic familiarity with tracing/observability concepts (helpful, not required)
FAQ
What is Langfuse Documentation — LLM Observability, Tracing and Evals?
The official documentation for Langfuse, an open-source AI engineering platform covering tracing, prompt management, evaluation and dashboards for LLM applications. It is for AI engineers who have an app in production (or heading there) and need to see what their agents actually did, what it cost, and whether quality is regressing.
Is Langfuse Documentation — LLM Observability, Tracing and Evals free?
Langfuse Documentation — LLM Observability, Tracing and Evals offers free content, with paid options for certificates or premium features.
What level is Langfuse Documentation — LLM Observability, Tracing and Evals for?
Langfuse Documentation — LLM Observability, Tracing and Evals is aimed at a intermediate audience. Recommended background: Python or TypeScript/JavaScript, An LLM application you can instrument, Basic familiarity with tracing/observability concepts (helpful, not required).
How long does Langfuse Documentation — LLM Observability, Tracing and Evals take?
Expect roughly ~2-3 hours to instrument a first app. Most learners work through it at their own pace.
What will I learn from Langfuse Documentation — LLM Observability, Tracing and Evals?
You'll learn: How to instrument an LLM app or agent so every span — prompt, tool call, retrieval, model response — is traced; How to debug a failing agent run by replaying its trace step by step with inputs, outputs and latency; How to attribute token cost and latency per user, session, feature or model; How to version, label and deploy prompts independently of application code; How to run LLM-as-a-judge and custom evaluators against datasets to catch quality regressions; How to instrument via OpenTelemetry or a first-party integration (OpenAI SDK, LangChain, LlamaIndex, LiteLLM).