home / platform / human-in-the-loop
Platform

Human in the loop: the AI prepares, you approve

Your Aivell does the work; you keep the judgment. It prepares each sensitive step, pauses, and waits for approval — a notification on your phone, answered in a tap. Choose the level per chore and per tool: get notified on every action, on specific tools, or approve each request individually.

Watch the demo

The pause

The scary version of AI automation is the one that acts first and explains later. An Aivell is built around the opposite move: the pause. It reads the invoice, checks the math, drafts the reply, fills the form — and then, if the step is sensitive, it stops. The work is done up to the point of commitment; the commitment stays yours.

You get a notification on your phone: who, what, how much. “Giulia · needs approval · Pay supplier invoice €1,240?” One tap to approve, one to reject. Approve and the chore resumes exactly where it paused. Reject and it follows the chore’s instructions for a no — which you also wrote.

Three levels of control

Trust isn’t binary, so the control level isn’t either. Per chore, per tool, you choose one of three:

Notify on everything. Your Aivell acts on its own and every action shows up in your feed. You watch the new colleague work without standing behind their chair.

Notify on specific tools. Routine tools run silently; the ones that matter — payments, anything leaving your domain, edits to the master sheet — ping you as they happen.

Approve each request. The hard stop. Chosen actions are prepared, then held until a human says yes. No approval, no action — not eventually, not by timeout, not ever.

Most teams run all three at once: watch the email triage, get pinged on customer-facing replies, hold every payment. The levels also tighten or relax over time — week one everything is held, month three only the money is. And they compose with guardrails: approval decides when an action runs, guardrails decide what actions are possible at all.

Judgment stays in the building

This is the honest division of labor. Your Aivell is better than anyone at reading two hundred pages without losing focus; you are better at knowing that this particular supplier gets a phone call, not an email. Human-in-the-loop keeps each side doing what it’s good at — the machine prepares, the human decides.

Every request, decision and timestamp lands in the audit trail on the box, so “who approved this?” always has an answer. The box stays in the office. You, demonstrably, don’t have to.

Questions, answered.

Do I have to approve everything it does? +

Only if you want to. You pick the control level per chore and per tool — from a running notification feed you can ignore, to a hard stop on every payment. Most people start strict and loosen it as the audit trail earns their trust.

What happens to the work while it waits for me? +

It stays parked. The step in question is prepared but not executed, and it cannot execute until someone says yes. Everything else keeps moving — other chores run, and unrelated steps of the same chore that don't need approval carry on.

Does approving from my phone mean my documents go through a cloud? +

No. The work, the documents and the models stay on the box in your office. Your phone is a remote control: you see the request and the context you need to decide, and your yes or no travels back — the data itself never moves.

Can approval rules live inside the chore itself? +

Yes, and that's the best place for them. "Over €2,000, ask first", "new supplier or changed bank details, pause and wait" — thresholds and exceptions are plain-language lines in the chore, so the pause happens exactly where your procedure says it should.

See it in practice.

What is an Aivell?

An Aivell is an on-premise AI colleague: a small box that plugs into your network, runs its own AI onboard and takes the repetitive work off your desk. Prompts, documents, data — nothing ever leaves your office. Unplug the internet: it keeps working.

You describe each task as a chore, in plain language. Your Aivell turns it into a solid, guarded procedure and runs it in the background — with guardrails on every tool, a complete audit trail, and approvals in your hand when you want them. One-time setup, fixed monthly fee. No tokens, no overages.

Watch the demo

onboard inference · air-gapped · fixed monthly fee