# Langfuse

> Open-source platform to trace, evaluate and improve LLM applications and agents

- Category: [Developer Tools](https://finderatlas.com/categories/developer-tools)
- Pricing: Freemium
- Website: https://langfuse.com
- Platforms: Web, API, CLI
- API: available
- Made by: Langfuse, founded 2023, Berlin, Germany
- Last updated: 2026-11-08
- FinderAtlas page: https://finderatlas.com/tools/langfuse

## Overview

Langfuse is an open-source LLM engineering platform for tracing, evaluating and improving AI applications and agents. It records hierarchical traces of model calls, tool invocations and retrieval steps, and lets teams filter them by user, session, cost, latency or metadata. Around that tracing layer it provides prompt management, evaluation (LLM-as-a-judge, heuristic checks and human annotation), datasets and experiments, a playground, and dashboards with alerts for cost and latency.

## Key features

- Tracing
- Evaluation
- Prompt Management
- Playground
- Datasets and Experiments
- Cost and Latency Dashboards
- Open-Source Self-Hosting
- APIs and MCP Servers

## Pricing

- **Hobby** ($0): Free cloud tier for individuals and early projects
- **Core** ($29/month): Paid tier for small production teams
- **Pro** ($199/month): Teams needing long retention and compliance reports
- **Enterprise** ($2,499/month): Large organizations with security and support requirements

## Best for

- Production tracing and debugging of LLM apps and agents, prompt versioning, and systematic evaluation on real data

## Who uses it

- AI Engineers
- Software Developers
- ML Engineers
- Product Teams
- Enterprises building LLM applications

## Strengths

- MIT-licensed core with free self-hosting, avoiding vendor lock-in.
- Covers tracing, prompt management and evaluation in one platform.
- Free Hobby tier with 50k units per month and no credit card required.
- Over 100 integrations, plus OpenTelemetry for languages without a native SDK.
- Core plan at $29/month with unlimited users.

## Limitations

- Hobby is limited to 2 users and 30 days of data access.
- SOC 2, ISO 27001 and HIPAA support start at the $199/month Pro plan, and SSO and RBAC need a $300/month add-on.
- Audit logs, SCIM and SLAs are restricted to the $2,499/month Enterprise plan.
- Usage-based unit overage adds cost for high-volume tracing.

## Not recommended for

- Non-technical users, or teams that want a fully managed tool with no instrumentation work

## Integrations

- LangChain
- Vercel AI SDK
- LiteLLM
- Pydantic AI
- Google ADK
- CrewAI
- LlamaIndex
- OpenAI Agents
- Mastra
- OpenAI
- Anthropic
- Amazon Bedrock
- Azure OpenAI
- Google Gemini
- Mistral AI
- vLLM
- Groq
- Claude Code
- Cursor
- Dify
- n8n
- PostHog
- OpenTelemetry

## FinderAtlas editorial scores (out of 10)

- Ease of Use: 7.8/10
- Feature Depth: 8.8/10
- Value for Money: 8.7/10
- Performance & Speed: 8.4/10
- Documentation & Support: 8.5/10

## Alternatives

- [Browserbase](https://finderatlas.com/tools/browserbase): Cloud browser infrastructure and APIs that let AI agents browse and act on the web ([compare](https://finderatlas.com/compare/browserbase-vs-langfuse))
- [Momentic](https://finderatlas.com/tools/momentic): AI QA platform that writes, runs and maintains end-to-end tests from plain English ([compare](https://finderatlas.com/compare/langfuse-vs-momentic))
- [Warp](https://finderatlas.com/tools/warp): Open-source terminal with built-in AI agents and cloud infrastructure for coding workflows ([compare](https://finderatlas.com/compare/langfuse-vs-warp))
- [fal](https://finderatlas.com/tools/fal): Run image, video, audio and 3D generative models through one fast developer API ([compare](https://finderatlas.com/compare/fal-vs-langfuse))
- [E2B](https://finderatlas.com/tools/e2b): Secure cloud sandboxes that give AI agents isolated Linux machines to run code ([compare](https://finderatlas.com/compare/e2b-vs-langfuse))
- [Replicate](https://finderatlas.com/tools/replicate): Run, fine-tune and deploy open-source AI models through a pay-per-use API ([compare](https://finderatlas.com/compare/langfuse-vs-replicate))
- [Browser Use](https://finderatlas.com/tools/browser-use): Open-source library and cloud platform for AI agents that control real web browsers ([compare](https://finderatlas.com/compare/browser-use-vs-langfuse))
- [OpenRouter](https://finderatlas.com/tools/openrouter): One API and one bill for hundreds of AI models across many providers ([compare](https://finderatlas.com/compare/langfuse-vs-openrouter))

## Recent updates

- 2026-01-01: ClickHouse announced it acquired Langfuse; Langfuse remains open source with Cloud continuing as a standalone service

## FAQ

**Is Langfuse open source?**

Yes. Langfuse is MIT licensed and free to self-host. ClickHouse said after acquiring the company that core features remain under the MIT license.

**Who owns Langfuse?**

ClickHouse announced the acquisition of Langfuse in January 2026, alongside its Series D round. ClickHouse says Langfuse Cloud keeps operating as a standalone service with the same SLAs and support.

**What counts as a unit on Langfuse Cloud pricing?**

Plans are metered in units of ingested data. Hobby includes 50k units per month, and Core, Pro and Enterprise include 100k, with overage billed per 100k units.

**Which languages and frameworks does Langfuse support?**

Python and TypeScript have native SDKs, and Go, Java, .NET, Ruby, PHP and Swift can send data via OpenTelemetry. Integrations include LangChain, LlamaIndex, Vercel AI SDK, LiteLLM, OpenAI and many others.

**Does Langfuse offer discounts?**

Yes. Early-stage startups get 50% off the first year, research and students can get up to 100% off with limits, non-profits get $199 in monthly credits, and open-source projects get $300 in monthly credits for the first year.

**Can I self-host Langfuse?**

Yes. Docker Compose is suggested for local setups, and Kubernetes (Helm) and Terraform templates cover production deployments on AWS, GCP and Azure.