Security and isolation
What the tenant boundary actually gives you, stated so you can check it rather than trust it.
The boundary
Your tenant is the outermost clamp. A token issued to your tenant sees your tenant's projects, runs, users, clusters, secrets and credentials — and nothing else. Another tenant's run id answers 404, not 403: a 403 would confirm that the id exists, which is itself a disclosure.
Instance-wide surfaces — minting tenants, listing every tenant, issuing signup invites — require an *instance administrator*, a role no tenant token carries. Those return 403 for you by design.
Things worth knowing precisely
`admin` scope means tenant admin. It is administrative over your tenant, not over the instance. The two are separate axes.
A machine token has no user. Some surfaces — knowledge in particular — are per-person and refuse a token with no owning user outright, with no administrative bypass. That is deliberate: an admin cannot read a colleague's personal collection by virtue of being an admin.
Registry credentials authenticate the registry only. They are written into a Docker config on every node that pulls, so a leaked one must stop at images rather than become an API token. The same confinement applies to git-mirror read credentials.
Secrets are one-way. Sealed values are never returned, only fingerprinted.
Second factor
Browser sign-in is GitHub OAuth plus TOTP. Enrolment happens on first login and is not optional. Store your recovery codes when they are shown — like the credentials above, they are displayed once.
What we do not claim
Hive holds no SOC 2, ISO 27001 or comparable certification, and has not been independently penetration-tested. It is a private-beta product and says so rather than implying otherwise. Judge the isolation claims above by testing them — they are stated in a form that makes that possible.
This page is the hive-docs knowledge collection your agent reads inside the product, published verbatim. Markdown version