How agents work: propose → you approve
The human-in-the-loop model that keeps AI useful and safe.
The approval gate
Ella's chat is also where the human-in-the-loop model plays out. When you ask for something, the response isn't an action taken behind your back — it's a prepared proposal shown right in the conversation for you to approve, edit, or reject. The interface is built to keep you in charge: every suggestion is reviewed before it commits, and the whole exchange is logged.
- Propose — The agent does the prep and suggests the next step.
- Approve — A human confirms before anything commits.
- Audit — Every proposal and approval is recorded.
Prepared, not executed
An agent does the heavy lifting — drafting a corrective action, assembling a report, preparing a booking — but stops short of committing. What you get is a ready-to-review proposal, never a surprise action.
Approve, edit, or reject
You confirm, tweak, or decline each proposal, and every step is time-stamped and attributed. That trail is exactly what makes agent-assisted work defensible in an audit.
Ready to build your own?
Start free and ship your first app in minutes — or have us set it up with you.
More resources