Hire it, brief it, set its limits

There is no bot to build and no workflow to draw. An employee arrives knowing its job. What it does not know yet is your business - and that is what you tell it.

Four steps, and it is at work

Choose the role

Support, reception, sales, orders and collections, bookkeeping or reporting. Each comes with the right skills, and each skill starts at a cautious level.

Answer a short interview

What you sell and to whom, which language and dialect, your opening hours, and who handles what it cannot. The configuration is written from your answers, and you can read all of it.

Give it your information

Your documents, your website, your price list. It answers from them and shows which one it used - and when the answer is not there, it says so and brings in someone from your team.

Decide how far it may go

For each skill: suggest only, draft for your approval, act and tell you, or act within a budget. Start everything at approval, and raise one skill once you have watched it for a week.

You set the limits. The system enforces them.

Each skill an employee has sits at one of four levels, and you choose which. The limit is enforced by the platform on every action - it is not a request the AI may ignore - and each action records the level that applied and why.

L0 · Suggests onlyPrepares the reply or the action, and leaves the doing to you.
L1 · Drafts for approvalPrepares the work and places it in your approvals list. Nothing happens until you say yes.
L2 · Acts, then tells youDoes it, then lets you know what it did.
L3 · Acts within a budgetGets on with it quietly, as long as it stays within the budget you set.

An employee's skills, each with its trust level chosen from a list and the level that actually applies shown beside it. — A level per skill, and the one that actually applies, with the reason.

Approve from wherever you are

Everything that needs a person waits in one list: what the employee wants to do, the order and the amount, why it is asking instead of acting, and three choices - approve, edit and approve, or decline. The same card works on your phone, and on WhatsApp if you would rather not open anything.

An approval card: the refund, the order number and amount, why it is waiting, the drafted details, and Approve, Edit and approve, and Reject. — What it will do, why it is asking, and your three choices.

The same approval on a phone. — On your phone, where most decisions get made.

Test it before customers meet it

Write down what a good conversation looks like - what a customer says, and what must be true afterwards - and run it. Each check passes or fails with the full conversation beside it. Change the employee after a passing run, and you are told the result is out of date.

Then see what it did

This is one row of the record, and it looks the same wherever the action appears - in a conversation, on a task, in the approvals list. One account of what happened, everywhere.

A WhatsApp conversation in the console: the customer's message, and the refund the employee drafted shown inline where it happened. — A conversation, with what the employee did shown where it did it.

The audit log: six actions, each with its trust level, the skill in plain words, the identifiers, its status and when it ran. — The audit log. Every action and the level it ran at - recorded before it ran.

Changes stay in draft until you publish

Customers keep talking to the version you published while you edit. Every version is kept, with who published it, what changed and why - and you can restore an earlier one in a click.

Start free for 14 days Book a demo