Skip to main content

Self-host Agenta

Agenta is an agent studio. When you self-host, you run the studio and the services that let you build, run, and evaluate agents on your own infrastructure.

What you run​

A self-hosted deployment is a set of containers:

  • Web and API serve the studio and its REST API.
  • Services runs workflows and routes agent runs to the runner.
  • The runner runs agent code inside a sandbox. See How agents run.
  • Postgres, Redis, and a durable store hold data, queues, and agent working directories.

Agents are the reason most people self-host Agenta, so agent execution is a first-class part of the deployment, not an add-on. If you only want to try Agenta locally, the Quick start gets you a running stack in a few minutes.

Editions​

Agenta ships in two editions:

  • OSS is the open-source edition, licensed under the MIT license. It covers the studio, the API, the runner, and local and Daytona sandbox execution. The OSS Docker Compose stack is the supported self-hosting path.
  • EE is the enterprise edition. It adds access control features such as SSO, roles, and dynamic access controls. EE is licensed separately.

Both editions self-host the same core studio. The pages in this section apply to OSS unless a page says otherwise.

Community-maintained deployments​

The OSS Docker Compose quick start is the default, supported self-hosting path.

Some deployment surfaces are community projects. Treat them as best-effort:

  • There is no expectation of official support.
  • Fixes and upgrades may depend on contributor availability.
  • Expect to troubleshoot and, in many cases, contribute back.

Today, the Railway OSS deployment is the community-maintained surface documented here. New community-maintained guides follow the same expectation. To improve one or propose a new one, see Community Projects.

Where to start​