AI Assistant

The assistant cannot answer from memory. It reaches your data through a typed registry of read-only tools — capacity overviews, rack and PDU detail, three-phase load, failover-overload checks, design-vs-measured reconciliation, energy rates, audit history, and search — and any figure it reports traces back to a specific call against your own records.

There is no path from a conversation to a change. Ask it to move, rename, or reconfigure anything and it declines in one sentence and hands you the screen where you can do it yourself. It cannot create, edit, or delete, and it will not claim to have done so.

The sidebar opens with starting questions drawn from what you are looking at — show me stranded capacity, why is this rack flagged, where could I land a new 2U server with 500W of headroom, what changed in the last 24 hours. Operators get a real answer before they have learned what to type.

The assistant runs as you. The same row-level tenant isolation and role permissions that govern the interface govern every tool call, so it can never surface a rack, tenant, or audit event you could not already open yourself.

The audit trail records that a user ran a query and when — never the question, the tool calls, or the answer. Compliance gets the access record it needs; your team gets a surface they can experiment on without their questions being read.

A per-user daily query allowance — 50 on Cloud Growth, 250 on Enterprise — resets at midnight UTC. No metered surprise on the invoice, and no way for one enthusiastic user to spend the month's budget in an afternoon.