Object storage

Two ways a bucket gets here, one screen for both:

• Managed — a MinIO the instance runs, as a container beside the databases. Where an app's uploads go without an account anywhere, and where you develop against S3 without paying for S3.
• Linked — an S3-compatible endpoint somewhere else, with keys you gave it: Amazon S3, Cloudflare R2, DigitalOcean Spaces, or any compatible server. Where a backup should go, because a backup kept on this machine is not one.

Both are an endpoint, a login and buckets inside it, and everything above the connection — browsing, uploading, attaching — is the same.

# there is no CLI noun for stores yet; the dashboard and the API are the doors

Every byte goes through the daemon

A download is read from the store and written to the response, an upload the other way. No cache, no copy, and no presigned URLs: a managed store's endpoint is a container name that resolves on the instance's network and nowhere else, so a link to it would be one only the daemon could follow.

Roles

Managing a store is an admin's, and so is everything inside one. A member can read a great deal about what the instance is wired to and never reads data — there is no way to see a row of a database from here either — and a bucket is data. A member's app still uses the store, because the app is given the keys.

MCP tools describe the storage and never its contents: an agent can see that backups/2026-09-01.dump exists and how big it is, and cannot open it, create a store or link one. It can attach an app, because that hands over no secret.

In the dashboard

Object storage in the sidebar, beside Databases. A store's page opens on Overview (charts and the endpoint), then Buckets, Apps, Logs; a linked store's Overview is the connection alone, since there is no container here to chart. Settings holds External access, Resources and Danger, and for a linked store the bucket it is pinned to.