Drafts, never decisions
What AI prepares arrives as a draft, a suggestion or a proposal. A person approves it, changes it or discards it. No model finalizes an invoice, sends a mail or books a cost.
AI native
Supplier invoices read from the PDF, a briefing that keeps itself current, an import that maps itself, and the assistant you already use, connected to your workspace. Every suggestion waits for a person.
Example from the app: a supplier invoice is captured, its fields are read from the PDF and the invoice waits as Needs review until a person approves it.
What AI prepares arrives as a draft, a suggestion or a proposal. A person approves it, changes it or discards it. No model finalizes an invoice, sends a mail or books a cost.
Each model call is recorded with its estimated cost, per workspace. Allowances cap what AI may spend, and the report shows where every call went.
Connect the tools you already use: Claude, ChatGPT or your own agent, through the MCP server with a key you can revoke. Your usage sits on an allowance you can see, and you decide what it may read and propose. You stay in control.
Supplier invoices
Drop a PDF or a photo into Supplier invoices. The fields are read from the document: supplier, number, date, net, VAT, total. The invoice waits as Needs review until a person checks and approves it, and possible duplicates are listed before anything is booked.
Briefings
Every project and company carries a short briefing: budget and hours, the next milestone, what is overdue, what was decided. It is rewritten after changes, so whoever opens the page on Monday reads the state of Friday evening, not of last month.
Example from the app: a project overview with its briefing, four sentences on budget, overdue tasks, the open invoice and what comes next, rewritten after the last change.
Import
Bring contacts, projects, tasks, time and expenses from a spreadsheet. The columns are matched to the right fields from the headers and a short sample. You check the mapping, see the preview and apply it; the same import run twice changes nothing.
Example from the app: the columns of a spreadsheet are matched to the right fields, five as AI suggestions, one chosen by hand, before the preview.
Your assistant
The MCP server gives the assistant you already use a key to your workspace: read what it may read, propose what it may change. With approval mode on, every proposed change lands in the inbox for a person to approve or reject. Finalizing and sending stay with people.
Example from the app: an assistant connected through the MCP server proposes a task; the proposal waits in the inbox until a person approves it, then it is executed.
Nothing runs up a bill in the background. Every model call is recorded, and the workspace report shows what AI did this month and what it cost.
You are not buying a model. You are buying a firm that watches the models for you, while your data protection and your approvals stay untouched.
Models move every few months. We follow them, test them against the way firms like yours bill, plan and write, and switch when it pays. Your team keeps the same screens and the same approvals.
Every suggestion is a draft. A person approves it, and no model finalizes an invoice, sends a mail or books a cost.
Your workspace strictly separated from every other, every AI provider named in the privacy notice, keys you revoke at once.
These features are in the product and run on the same models. They are switched on step by step; ask in the discovery call where they stand.
Specified, not built. Named here so you know where the product is going.
Thirty minutes on where AI saves your team time today, and where it should stay out of the way.
No slides. No sales pitch. A genuine review.