Langfuse Documentation — LLM Observability, Tracing and Evals
by Langfuse
Trace, evaluate and version-control your LLM app with an open-source platform you can self-host.
Overview
Langfuse is an open-source LLM engineering platform — tracing, prompt management, evaluation, datasets and an interactive playground — and this is its official documentation. The MIT-licensed core (with a separate enterprise folder carved out) has over 33,000 GitHub stars, and the company was acquired by ClickHouse in January 2026, announced alongside ClickHouse's $400M Series D; both the managed cloud and self-hosted deployments continue. The docs are organised around three product pillars. Observability covers the trace data model — traces containing nested observations, spans and generations that record prompts, completions, token counts, cost and latency — plus sessions and users for multi-turn conversations, agent-graph visualisation, and custom dashboards and alerts. Prompt Management covers versioned prompts fetched at runtime, labels for staged rollout, and the Playground for iterating interactively. Evaluation covers LLM-as-a-judge evaluators, code evaluators, user feedback, manual annotation queues, datasets and dataset experiments for regression testing. Instrumentation is documented for the Python and JS/TS SDKs, drop-in wrappers for the OpenAI SDK, integrations for LangChain, LlamaIndex and 100+ other libraries, and native OpenTelemetry ingestion so traces can arrive without a Langfuse-specific SDK. Self-hosting guides cover Docker Compose, VMs, Kubernetes via Helm, and Terraform modules for AWS, Azure and GCP. Langfuse Cloud is metered in units: Hobby is free with 50k units per month and 30 days of data access, Core is $29/month, Pro $199/month and Enterprise $2,499/month, each including 100k units with overage from $8 per 100k.
At a Glance
- Topic
- Frameworks
- Level
- Intermediate
- Format
- Documentation
- Cost
- Freemium
- Duration
- ~3-5 hours for the core observability, prompt and evaluation guides; reference thereafter, self-paced
- Provider
- Langfuse
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Instrument an LLM app so every model call, tool and retrieval step lands in one trace
- ✓Model multi-turn conversations with sessions and attribute traces to individual end users
- ✓Move prompts out of the codebase into versioned, labelled prompt management fetched at runtime
- ✓Build LLM-as-a-judge and code-based evaluators and wire them into annotation queues
- ✓Create datasets and run dataset experiments to catch regressions before shipping a prompt change
- ✓Track token usage, cost and latency per model and surface them on custom dashboards
- ✓Self-host Langfuse with Docker Compose, Helm or Terraform on AWS, Azure or GCP
- ✓Send traces over OpenTelemetry instead of a vendor SDK to keep instrumentation portable
Highlights
- •MIT-licensed core with 33k+ GitHub stars — the whole platform can be self-hosted for free
- •Native OpenTelemetry ingestion, so your instrumentation survives a later change of vendor
- •Docs cover the awkward parts: cost attribution, data retention windows and multi-tenant self-hosting
- •Acquired by ClickHouse in January 2026; cloud and self-hosted both continue, but weigh the ownership change in a long-term bet
- •Pricing is documented in concrete units and dollar figures rather than a 'contact sales' wall
Who It’s For
Best For
- ✓Engineers taking an LLM prototype into production and needing real traces
- ✓Teams standing up evaluations and regression tests for prompts and agents
- ✓Platform teams choosing a self-hostable alternative to closed observability SaaS
Prerequisites
- •Working knowledge of Python or TypeScript and an existing LLM application to instrument
- •Basic familiarity with tracing concepts such as spans, traces and structured logs
- •Docker and Kubernetes basics if you intend to follow the self-hosting guides
FAQ
What is Langfuse Documentation — LLM Observability, Tracing and Evals?
The official documentation for Langfuse, the open-source LLM observability and evaluation platform now owned by ClickHouse. It covers instrumenting an application with traces, spans and generations, managing versioned prompts outside your codebase, and building LLM-as-a-judge and dataset-based evals. Includes Python and JS/TS SDKs, OpenTelemetry ingestion, and full self-hosting guides for Docker, Kubernetes and Terraform.
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: Working knowledge of Python or TypeScript and an existing LLM application to instrument, Basic familiarity with tracing concepts such as spans, traces and structured logs, Docker and Kubernetes basics if you intend to follow the self-hosting guides.
How long does Langfuse Documentation — LLM Observability, Tracing and Evals take?
Expect roughly ~3-5 hours for the core observability, prompt and evaluation guides; reference thereafter, self-paced. Most learners work through it at their own pace.
What will I learn from Langfuse Documentation — LLM Observability, Tracing and Evals?
You'll learn: Instrument an LLM app so every model call, tool and retrieval step lands in one trace; Model multi-turn conversations with sessions and attribute traces to individual end users; Move prompts out of the codebase into versioned, labelled prompt management fetched at runtime; Build LLM-as-a-judge and code-based evaluators and wire them into annotation queues; Create datasets and run dataset experiments to catch regressions before shipping a prompt change; Track token usage, cost and latency per model and surface them on custom dashboards; Self-host Langfuse with Docker Compose, Helm or Terraform on AWS, Azure or GCP; Send traces over OpenTelemetry instead of a vendor SDK to keep instrumentation portable.
Topics
Sources
This page was written from 5 sources, 2 on domains other than langfuse.com.