AI Mode and Assistant

AI Mode is SHM's role-aware conversational workspace. It can discover and use the public API available to the signed-in root, reseller, or account user without creating a parallel permission system.

Where to find it: Server > AI for root and resellers; Account > AI Assistant for the selected account when the AI feature is enabled.

Available to: Root for the server workspace, resellers for owned-account workflows, and account users for their own account when AI is enabled by policy.

What this page does

Use AI Mode to translate a plain-language request into current SHM reads, diagnostics, plans, and supported actions. It queries the live role-filtered API catalog instead of relying on a small hard-coded list of favorite endpoints.

Before you start

Confirm the active role and selected account. Do not paste secrets into chat. For any mutation, review the target, parameters, expected result, and recovery path; friendly conversation does not make a destructive call less destructive.

Verified API coverage

Intentional exclusions

• AI provider keys: chat cannot read, create, or delete provider credentials.
• Interactive terminal sessions: chat cannot mint a browser terminal session.
• WHMCS login tickets: account chat cannot create an external login ticket.
• Private routes: /v1/internal/... remains orchestration-only and never appears in the public AI catalog.

How endpoint discovery works

1. AI calls GET /v1/ai/catalog with its authenticated role, surface, and authorized account.
2. SHM filters private routes, role mismatches, unowned accounts, and intentionally blocked credential/session operations.
3. The result includes the real method, path template, purpose, every parameter, accepted values, derivation notes, and a safety policy.
4. Reads may run when values are ready. Mutations remain catalog-only until the workflow and confirmation policy allow execution.

Good requests

• “Check why example.com returns a 502 and show the evidence.”
• “List the DNSSEC state first, then explain what enabling it will require at the registrar.”
• “Show every parameter you would send before changing this mailbox.”
• “Prepare a storage-cleanup preview; do not apply it.”

Proof, not guesswork

AI should derive current ids from list or context endpoints, run bounded diagnostics, and verify a changed state with a follow-up read. If it cannot establish ownership, current state, or a required value, the correct result is a precise blocker—not a confidently invented resource.

How to use it

1. Open the correct AI surface.
2. Name the account, domain, mailbox, database, service, or job precisely.
3. Ask for current evidence and the planned endpoint before a write.
4. Review every parameter and confirmation.
5. Verify the outcome in the normal SHM page and the customer-facing service.

Result and next check

A useful session ends with source-backed state, the correct role-scoped endpoint and complete parameters, and a separately verified outcome. The assistant may save clicks; it does not get a skeleton key.