What does each AI tool see?
Only the rows your role allows, scoped to your tenant. Permissions are enforced once at the substrate boundary, not per-tool, so a new AI client inherits exactly the same view a new employee with the same role would get.
Is everything audited?
Yes — every tool call is hash-chained into a tamper-evident log with the calling user, the tool, the arguments, the policy decision, and the result digest. Same chain whether the call came from Claude, ChatGPT, the CLI, or a custom agent.
Does the URL change as I scale?
No. The URL is permanent. When a tenant promotes from the shared pool to a dedicated node, the routing is a database column change — your AI clients keep the same URL forever.
Where do tokens live?
In each AI client. We never see them after issuance — the OAuth flow runs through your browser. Revoke from the client (Claude.ai / ChatGPT) or contact us to rotate the signing key (kills all tokens at once).