Langfuse
FreemiumOpen source agent evals and observability.
Published 1 October 2026
Scores
Popularity3/5
Around 34K GitHub stars and the most-cited open-source option in LLM observability comparisons, well known to teams shipping LLM features but a young category that few developers outside AI work have looked at yet.
Learning Curve2/5
Framework integrations add tracing with a callback handler or a decorator, and the cloud version needs only an API key. Evaluations and datasets take more thought but are optional.
Flexibility5/5
OpenTelemetry-based instrumentation, SDKs in Python and TypeScript, integrations for most agent frameworks, a public API, and a self-hostable codebase let teams wire it into nearly any stack.
Performance4/5
The ClickHouse-backed analytics layer handles high trace volumes and fast dashboard queries, and SDKs send traces asynchronously so they add little latency to the application.
Portability5/5
MIT-licensed core, self-hosting on any infrastructure, and OpenTelemetry instrumentation mean traces and prompts can move between the cloud and self-hosted deployments or to another OTel backend.
About Langfuse
Langfuse is an open-source LLM engineering platform for seeing what LLM applications and agents actually do in production. It records every model call, tool call, retrieval step, and agent turn as nested traces, then layers cost, latency, and quality analytics on top so teams can find slow, expensive, or wrong runs and understand why.
Beyond tracing it covers the rest of the improvement loop. Prompt management versions prompts outside the codebase so they can be changed and rolled back without a deploy, with a playground for testing them against models. Evaluations score outputs with LLM-as-a-judge, custom code, user feedback, or manual annotation queues, and datasets turn real traces into regression tests for new prompts or models. Metrics dashboards track usage, spend, and scores over time by user, session, or feature.
Instrumentation is framework-agnostic and built on OpenTelemetry. Native integrations cover the OpenAI SDK, LangChain and LangGraph, LlamaIndex, the Vercel AI SDK, Pydantic AI, CrewAI, LiteLLM, and most other agent frameworks, plus Python and TypeScript SDKs for custom code. That makes it a common neutral choice for teams that don't want their observability tied to one framework vendor.
The core is MIT-licensed and can be self-hosted for free with Docker Compose or Kubernetes, with a few enterprise features under a separate licence. Self-hosting at scale means running its ClickHouse, Postgres, Redis, and object-storage dependencies. Langfuse Cloud is the managed option, billed by usage units (traces, observations, and scores) with a free Hobby tier. ClickHouse acquired Langfuse in January 2026; the licence and pricing model were kept as they were.
Key Features
- Nested tracing of LLM calls, tool calls, retrieval, and agent steps
- Cost, latency, and token analytics per user, session, and feature
- Versioned prompt management with a model playground
- LLM-as-a-judge, code-based, and human annotation evaluations
- Datasets built from production traces for regression testing
- OpenTelemetry-based SDKs with integrations for LangChain, LlamaIndex, and the OpenAI SDK
- Self-hostable with Docker Compose or Kubernetes
Pros
- Genuinely open source and self-hostable without an enterprise contract, unlike LangSmith
- Works with almost any framework or model provider, so it survives a stack change
- Free Hobby tier and a $29 Core plan make the cloud version cheap to start
- Tracing, prompts, and evals live in one tool instead of three
Cons
- Self-hosting at scale means operating ClickHouse, Postgres, Redis, and blob storage
- Usage-unit billing grows quickly for agents that emit many observations per run
- SSO and SCIM sit behind paid add-ons or the Enterprise plan
- Deep LangGraph-specific debugging views are less polished than LangSmith's
Langfuse Pricing
Freemium- · MIT-licensed core with no usage limits
- · Runs on Docker Compose or Kubernetes
- · 50,000 units per month on Langfuse Cloud
- · Two users and 30-day data retention
- · 100,000 units per month, then $8 per 100,000
- · Unlimited users and 90-day retention
- · Three-year data retention and higher rate limits
- · SOC 2 and ISO 27001 reports
- · Optional Teams add-on for SSO
- · SCIM, audit logs, and SLAs
- · Dedicated support
Tools Related to Langfuse
Integrates with Langfuse(3)
LiteLLM has a built-in Langfuse callback, so a gateway can log every routed request, with tokens and cost, to Langfuse by setting it as a success callback.
Langfuse ships a LangChain callback handler that records chains, tool calls, and LangGraph runs as nested traces, an open-source option for LangChain teams that want to self-host their observability.
Langfuse traces LlamaIndex applications through an official integration, recording each retrieval, LLM call, and agent step with its cost and latency.
Alternatives to Langfuse(1)
Langfuse is open source and free to self-host, with a framework-neutral design; LangSmith is proprietary and cloud-first, with the deepest debugging views for LangChain and LangGraph and managed agent deployment on top.
Vendor
ClickHouse
Website →Tags
Details
- Maintained
- Yes
- Monitoring
- Llm Tracing, Evaluations, Metrics
- Hosting
- Cloud & Self-hosted
- Open source
- Yes
- GitHub stars
- 34.4k
- Stars updated
- 2026-09-30