Langfuse

Freemium

Open-source LLM observability and prompt management platform. MIT-licensed with a free self-host that includes all core features. Strong prompt-versioning model and a mature observation data structure — traces, observations (generations/spans/events), sessions, scores. Acquired by Clickhouse in January 2026; open-source code remains actively maintained.

Visit website GitHub

Pricing

Freemium

Type

Automation

Languages

Python, TypeScript

// VERDICT

Reach for Langfuse when you want open-source, self-hostable tracing plus evaluation and prompt management for LLM apps. Skip it when you prefer a fully managed vendor platform or only need code-only evals.

Best for

Open-source LLM engineering platform - tracing, evaluation, prompt management and metrics for LLM apps, self-hostable, framework-agnostic.

Avoid when

You want a fully managed-only vendor, or a lightweight code-only eval without a platform.

CI/CD fit

SDK tracing · self-host or cloud · eval/score APIs

Languages

Python · TypeScript

Team fit

LLM app teams · Teams wanting open-source observability · Dev/QA debugging + evaluating

Setup

Medium

Maintenance

Low

Learning

Intermediate

Licence

Freemium

// BEST FOR

Open-source, self-hostable LLM tracing and observability
Evaluation and scoring of outputs
Prompt management and versioning
Metrics on cost, latency and quality
Framework-agnostic instrumentation
Debugging chains/agents from full traces

// AVOID WHEN

You want a fully managed-only vendor
A lightweight code-only eval is enough
You won't run/host the platform (though cloud exists)
You're not building LLM apps
Only prompt comparison is needed
Turnkey enterprise support is essential

// QUICK START

# self-host via Docker, or use Langfuse Cloud
pip install langfuse   # or npm i langfuse
# instrument your app -> traces, prompts, evals

// ALTERNATIVES TO CONSIDER

Tool	Choose it when
LangSmith	You want a managed platform tied to LangChain.
Arize Phoenix	You want OpenTelemetry-based open-source observability.
Laminar	You want another open-source trace + eval option.

// FEATURES

First-class prompt management with versioning, A/B testing, and rollback
Typed observation model (generations, spans, events) for structured LLM tracing
Mature eval harness with dataset workflows and LLM-as-judge scoring
OpenTelemetry-native — works as a generic OTLP backend if you already instrument with OpenLLMetry
Self-host via Docker Compose or managed Cloud — same feature set both ways

// PROS

Genuinely free self-hosted — no seat tiers, no feature gating on the OSS image
Largest open-source community in the space — 20K+ GitHub stars, 12M+ monthly SDK downloads
Permissive MIT licence — no vendor lock-in even if you start on managed Cloud

// CONS

Self-host requires PostgreSQL, ClickHouse, Redis, and S3 — significant infrastructure for small teams
Trace UX is observation-tree-first; reading a 2,000-span agent run is slower than transcript-based tools
Acquired by Clickhouse in early 2026 — long-term roadmap is now subject to Clickhouse's commercial direction