Control & confidence

Useful AI. Visible boundaries.

Understand the intended controls, your responsibilities and the questions to ask before production.

Designed-in controls

Design requirements—not a certification claim.

Workspace boundaries

Permissions and membership scope govern reads, files, searches, exports and agent actions.
Review consequential actions

A changed amount, recipient or record version requires a refreshed preview and the required approval.
Separate public and private

Marketing and public search never need private expense records or conversations.
Keep a useful audit trail

Confirmed actions are recorded with the same history as manual ones.
Treat documents as data

Text inside a receipt or email is evidence to read, not an instruction the assistant obeys.
Provider choices are explicit

AI providers are chosen deliberately, and what is sent to them is limited to what the task needs.

Per-workspace permissions

The agent works inside the current person’s permissions in the current workspace.
Preview and confirmation

It shows a plan with affected records. Nothing changes until a person confirms.
No approving or paying

It cannot approve or pay by itself. Those keep their own controls.
Usage allowances

AI use is limited by the plan’s published allowance, shown to you.
Manual fallback

When AI is unavailable or the allowance runs out, the manual tools keep working.
Estimates are labelled

Results can be wrong, so source records are shown for you to check.

Can the assistant change things on its own?

No. It prepares a preview and a person with permission confirms.

Does AI widen what I can access?

No. An instruction cannot expand what the current user is allowed to do.

Do I have to use it?

No. AI, Hybrid and Classic are views of the same records.

Can it make mistakes?

Yes, like any automated tool. That is why you review sources and previews before applying anything.

The demo shows a proposal, its scope and the outcome, with sample data.