Assistant

AI that proposes. People who decide.

The Assistant works inside your instance, on your records, with your permissions. It saves nothing that a person has not approved.

How it works

The Assistant does the typing. People make the decisions.

Ask in plain language: What you are looking at travels with your message, so "raise a CAPA from this" just works.
Proposals, not surprises: The Assistant holds its edits unsaved in a Proposal. You approve or reject the whole set.
Send it on: Not yours to decide? Send the Proposal to a person who may make every edit in it.
Marked in the history: Approved edits are your edits, marked as made via the Assistant in every record's history.

A private thread with the Assistant.

Private to you: Each Conversation belongs to one person. Nobody else reads it.
Picks up where you left off: Open the Assistant and your last Conversation resumes. After eight idle hours, a new one starts.
Your permissions: The Assistant sees and changes only what you could see and change yourself.

Recurring work, done in the background.

Recurring runs: Ask the Assistant to do a job at recurring times — a weekly review of overdue actions, for example.
A named approver: A Schedule names who decides its Proposals: a person or a role.
Standing approvals: Approve narrow kinds of edits in advance, so a Schedule can make them without asking. Never wider than its scopes.

Bring the AI tool you already use.

Through MCP: Connect Claude or another MCP-capable tool to your instance.
Exactly your roles: An AI Connection acts as you, with your roles — nothing more — until you revoke it.
Same approval rule: Like the Assistant, an AI Connection saves nothing until you approve its Proposal.
Your instance decides: An administrator allows AI Connections for the whole instance, or does not.

Tell us what you build and where it hurts today. We will walk you through the platform with your records, not a generic script.