Stay found
AI assistant
The editable brain your support assistant runs on — personality as versioned database rows — and a metered, tool-checked chat surface that answers from your live data.
The problem
A chatbot you can't retune without a deploy, that invents numbers and can't say who changed its behaviour, is a liability wearing a helpful face.
How it works
The assistant's personality is database rows an admin edits — with versions, one-click rollback, and a change log — so tuning how it behaves is a staff action, not a code change. Per-tenant overlays let one customer bend it without forking the rest, and a publish is refused if it deletes a required phrase, so an edit can't ship a bot that forgot its own name.
The chat surface reaches live data only through a whitelisted, permission-checked tool registry: it names a procedure, the server owns every parameter, and it can never read what the asker's own role can't. It's built on the metering rails, so every request lands in a ledger with token counts and cost — recorded even when the reader closes the tab mid-answer.
Both are live in the admin today — the editable brain and the chat surface that runs on it. This page shows the metering ledger behind it, honestly labelled a preview.
- Personality as versioned, revertible rows — no deploy to retune
- A publish that strips a required phrase is refused, not shipped
- Every behaviour change leaves a change-log row: who, when, what version
- Built on permission-scoped, metered rails; cross-tenant reach is structurally impossible
npm install @adminigloo/assistantBook a callAll features