Langfuse
langfuse.com
- Free plan
- Open source
- Linux
- Web
- API
- Self-hosted
What Langfuse is
Langfuse is an open-source platform for engineering teams that build AI agent applications. It brings debugging, analysis, and iteration tools together, with traces capturing prompts, model responses, token use, latency, tool calls, retrieval steps, and metadata. Prompt management covers versioning, deployment, testing, and links between prompts and traces. Evaluation options include LLM-as-a-judge, code evaluators, user feedback, manual labeling, datasets, and experiments.
Teams can connect through Python and JS/TS SDKs, OpenTelemetry, an API, CLI, MCP Server, and integrations with AI frameworks. The MIT-licensed core can be self-hosted with Docker, including on Docker Compose, Kubernetes, AWS, or Azure. Langfuse reports security audits, penetration tests, GDPR compliance, and a HIPAA-ready region; its cloud service uses TLS 1.2+ in transit and AES-256 for stored data. A free plan is available; paid plans start at $29/mo.
Who it suits
Langfuse suits teams building AI agent applications that need to trace activity, manage prompts, and evaluate outputs. Self-hosting and cloud options support different deployment preferences.
Good to know, and what to watch
- Traces prompts, tokens, latency, tools, and retrieval steps
- Links prompt versions to traces
- Includes multiple evaluation methods
- MIT-licensed core can be self-hosted
- Free plan includes 50k units monthly
- Free plan has 30 days of data access
- Free plan is limited to two users
- Core includes 90 days of data access
Plans
Every plan and priceHobby
Core
Pro
Teams Add-on
Free plan
Getting started
- Open https://langfuse.com/
- Choose the hosted web service or plan a self-hosted deployment
- Connect through the API, Python or JS/TS SDK, OpenTelemetry, CLI, or MCP Server
- Deploy the core with Docker Compose, Kubernetes, AWS, or Azure when self-hosting
- Select the Hobby free plan or a paid monthly plan
Limits to check
Hobby provides 50k units per month, 30 days of data access, two users, and community support through GitHub. Core includes 100k units monthly, 90 days of data access, and charges $8 per additional 100k units.
The verdict
Langfuse combines AI application tracing with prompt management and evaluation in one platform. Teams should weigh its free-plan limits and paid plan costs against their data-access and support needs.
Questions people ask
Is Langfuse open source?
The core is MIT-licensed and available for self-hosting.
Does Langfuse have a free plan?
Yes. Hobby costs nothing and includes 50k units per month.
What does Core cost?
Core costs $29 a month, billed monthly, with 100k units included.
Which deployment options are available?
You can use the web service or API, or self-host with Docker Compose, Kubernetes, AWS, or Azure.
Does Langfuse track token costs?
Yes. The platform provides cost tracking and token cost tracking.
What the maker publishes
- Access control
- Langfuse supports role-based access control and enterprise single sign-on through OIDC. source
- Acquisition news
- ClickHouse announced on January 16, 2026 that it had acquired Langfuse alongside a $400 million Series D financing. source
- API
- The public data platform includes APIs for observations, experiments, metrics, prompts, datasets, projects, exports, and an OpenAPI specification. source
- Cloud regions
- Langfuse Cloud lists US/HIPAA hosting in AWS us-west-2, EU hosting in eu-west-1, and Japan hosting in ap-northeast-1. source
- Company legal identity
- The legal entity is Langfuse GmbH, registered in Berlin, Germany, with an address on Oranienburgerstraße 91. source
- Company locations and size
- LinkedIn lists Langfuse as privately held, headquartered in San Francisco, founded in 2022, with 11–50 employees and an additional Berlin location. source
- Company status
- Langfuse joined ClickHouse in January 2026. source
- Documentation
- Documentation covers observability, prompt management, evaluation, metrics, APIs, SDKs, integrations, compatibility, deployment, and security with versioned self-hosting guides. source
- Encryption
- Langfuse Cloud protects traffic with TLS 1.2+ and stored data with AES-256 encryption. source
- Evaluation
- Evaluation features include LLM-as-a-judge, code evaluators, user feedback, manual labeling, datasets, and experiments. source
- Evaluations
- Evaluation supports LLM-as-a-Judge, heuristic or code evaluators, human review, annotation queues, and scoring of production data or experiments. source
- Experiments
- The UI can run prompt or model experiments over datasets and compare results side by side, optionally applying evaluators. source
- Extensibility
- Langfuse provides a CLI, MCP server, agent skill, REST APIs, query SDKs, and blob-storage exports for scripts, CI/CD, and coding-agent workflows. source
- Founded and accelerator
- Y Combinator says Langfuse began during its W23 batch in San Francisco while GPT-4 was initially released. source
- Funding
- Crunchbase lists one seed funding round with five investors, including Y Combinator, La Famiglia, and Lightspeed Venture Partners, but hides the total amount. source
- Headquarters
- Berlin, Germany; San Francisco, United States source
- Integrations
- Langfuse supports Python and JS/TS SDKs, OpenTelemetry, API, CLI, MCP Server, and integrations with popular AI frameworks. source
- Integrations
- Langfuse lists more than 100 integrations spanning OpenTelemetry, Python and TypeScript, model providers, agent frameworks, gateways, and developer tools. source
- Licence
- Langfuse states that all product features are MIT licensed, while a separate Enterprise license covers governance features such as project RBAC, audit logs, SCIM, and retention policies. source
- Limit: G2 performance
- The sole G2 reviewer reported that the product had recently become laggy and data sometimes failed to load. source
- Limit: Gartner scale and polish
- A Gartner reviewer said bulk queries can remain slow and that some workflows feel less polished as complexity and traffic increase. source
- Limit: independent practitioner reviews
- Enterprise DNA characterized Langfuse as engineer-oriented, requiring infrastructure capability and still having gaps in evaluation depth, UI polish, and compliance coverage. source
- Limit: self-hosting complexity
- Panoply said current self-hosting requires Postgres, ClickHouse, Redis, and S3, and reported slower UI behavior on heavy traces. source
- Metrics
- Metrics cover quality, cost, latency, and volume, with breakdowns by user, session, geography, feature, model, and prompt version. source
Checked against the maker's pages on 15 Oct 2026.