Langfuse is an open source LLM observability platform you can self-host — a LangSmith alternative that traces, evaluates, and debugs AI applications by capturing prompts, model calls, retrieval steps, and agent actions in one place.

What is Langfuse?

Langfuse is an open source LLM observability platform that traces, evaluates, and debugs AI applications. You instrument your app to capture every LLM call, retrieval step, and agent action as a structured trace, then layer on prompt management, evaluations, and datasets — so you can see exactly what your model did, why it cost what it did, and where it went wrong.

What is Langfuse best for?

Teams building LLM apps and agents who need production visibility — debugging multi-step chains, tracking token cost and latency, scoring output quality, and iterating on prompts — without routing sensitive traces through a hosted-only service. It fits engineers who already use frameworks like LangChain, LlamaIndex, or LiteLLM and want framework-agnostic tracing they can self-host.

What can Langfuse do?

Trace LLM calls, retrieval, embeddings, and agent steps as nested, inspectable spans
Manage and version prompts centrally, with strong caching so iteration adds no latency
Run evaluations — LLM-as-a-judge, code-based checks, user feedback, and manual labeling
Build datasets and test sets to benchmark changes before you ship them
Test prompts and model configs interactively in a Playground
Integrate with OpenAI, LangChain, LlamaIndex, LiteLLM, the Vercel AI SDK, and 20+ others
Instrument anything via OpenAPI, Python and JS/TS SDKs, and OpenTelemetry

Is Langfuse free?

Yes — Langfuse is free to self-host under the MIT license, and you only pay for your own infrastructure. Langfuse Cloud has a free Hobby tier (50k units/month) plus paid plans: Core at $29/mo, Pro at $199/mo, and Enterprise at $2,499/mo, priced on usage and data retention. You’re paying for managed hosting and enterprise features, not the core software.

Where does Langfuse fall short?

Self-hosting is infrastructure-heavy: a production deployment needs PostgreSQL, ClickHouse, Redis, and S3-compatible storage, which is far more than a single container.
It’s open core, not fully open. The MIT license excludes the ee folders, so features like SCIM provisioning, audit logs, and some SSO enforcement are gated behind paid or enterprise tiers.
It’s purpose-built for LLM apps, not a general-purpose APM. If you want full-stack infrastructure monitoring alongside LLM tracing, Datadog covers far more ground.

What does Langfuse replace?

Langfuse is a self-hosted alternative to LangSmith, LangChain’s hosted observability and evaluation platform, and it covers the LLM-tracing slice of Datadog. It does the same trace-evaluate-debug job for AI apps, but runs on your own infrastructure and is free at its core.

FAQ

Is Langfuse open source? Yes — the core is MIT licensed and free to self-host. It’s open core, though: enterprise features in the ee folders (SCIM, audit logs, some SSO controls) are licensed separately.

Can I self-host Langfuse for free? Yes. The MIT core is free to run; you only pay for your servers. Note it expects PostgreSQL, ClickHouse, Redis, and S3-compatible storage for a real production setup.

Is Langfuse a good LangSmith alternative? For teams that want to self-host and avoid per-seat pricing, yes — it offers comparable tracing, prompt management, and evaluations, and it’s framework-agnostic rather than tied to LangChain.

What do I need to run Langfuse? For local testing, Docker Compose gets you running in minutes. For production, plan for PostgreSQL, ClickHouse, Redis, and S3-compatible object storage, typically on Kubernetes.