Langfuse is an open-source AI engineering platform that helps teams build, monitor, and improve their LLM applications. It covers the full development lifecycle with tracing, prompt management, evaluations, and analytics dashboards, plus an optional AI Gateway that routes your AI calls through Langfuse and traces each one with usage and cost. Many teams start with tracing alone and adopt other features as they need them. Langfuse is used by 50,000+ companies and processes billions of observations per month. You can try it instantly with the public demo project or sign up for free.

Langfuse is an open-source AI engineering platform for tracing, evaluations, prompt management, and experiments, with an optional AI Gateway. Compared with LangSmith, Braintrust, Arize, Galileo, and Datadog Agent Observability, it is MIT licensed, OpenTelemetry native, self-hostable at the same scale as Langfuse Cloud, and model- and framework-agnostic. On a representative workload, Langfuse Cloud is cheaper than LangSmith, Braintrust, Arize, and Datadog Agent Observability. See Why Langfuse? for how we position the product, or the comparison pages for product-by-product differences.

Langfuse is available as a managed cloud service in global regions, or you can self-host it on your own infrastructure using Docker Compose, Kubernetes (Helm), or Terraform templates for AWS, GCP, and Azure. Self-hosting is free under the MIT license and includes all core features, such as observability, evaluation, prompt management, and datasets, with unlimited usage. The AI Gateway is fully available when self-hosting as well. You only pay for your own infrastructure. The self-hosted Enterprise plan adds project-level RBAC, audit logs, data retention policies, management APIs, and dedicated support with SLAs.

No. The AI Gateway is optional. To use it, swap the base URL and API key in your AI calls for the gateway's, and every call is traced without an SDK. Otherwise, the SDKs and OpenTelemetry integrations work as before. You can also combine both, so gateway calls show up inside your existing traces.

Tracing with the SDKs and OpenTelemetry is async and doesn't block your application. Data is sent in the background with automatic batching and queuing. The AI Gateway sits in the request path between your application and the provider, and is built in Rust and optimized for low latency. Prompts are cached client-side in the SDKs, so there's no latency impact after the first fetch, and guaranteed availability keeps your application working even if Langfuse is unreachable.

Yes. Langfuse Cloud is SOC 2 Type II certified, ISO 27001 compliant, GDPR compliant, and offers a HIPAA-ready region. Data is encrypted at rest and in transit, and you can choose between US, EU, Japan, and HIPAA data regions. For full control, you can self-host Langfuse on your own infrastructure with data masking and your own encryption keys. See our security overview for details.

Langfuse Cloud has a free Hobby plan with no credit card required. Paid plans use usage-based graduated pricing based on billable units, which are the traces, observations, and scores you send to the platform. Volume discounts apply automatically as you scale. You can also self-host Langfuse for free under the MIT license. See the pricing page for the full calculator and plan comparison.