Proof · Case study
vocabotics Console: the system we run our own company on.
A plain-English operating system for a small company. AI staff do the routine work. A person says OK before anything leaves the company. We built it, and we use it every day.
Our own system, used every day to run vocabotics.
As of 19 November 2026. Not on sale yet.
What is in it
Six facts you can ask us to show you.
Each one is a way the system is checked, not a claim about how clever it is.
About 4,650 automated tests
Run as CI gates on both Windows and Linux. A change that fails on either does not ship.
A person approves every outbound action
AI staff draft. Nothing leaves the company until a person has read it and tapped Approve.
Gmail through keyless Workspace delegation
Console reaches mail through Google Workspace delegation, not a stored key file. Jon still approves each email.
Buyer-intent Radar
Watches public sources for organisations showing signs of wanting what we build, and scores them for a person to review.
An audit trail
Every draft, approval and send is recorded with who, what and when, so any action can be traced afterwards.
An AI gateway with caps
Every AI request goes through one gateway with spending caps. When a cap is reached, the request is refused, not billed.
The approval flow
Autonomous inside the building. An OK at the door.
This is how an outreach email moves through Console. Every step is recorded, and the send step cannot run until a person has approved.
A drawing of the flow, not a screenshot: the real screens sit behind a sign-in.
What this does not prove
Where to be careful.
- This is our own company on our own system. It shows how we build and check software, not how a client's project will go.
- A test count says the checks exist and pass. It does not say the product is useful. That is judged by daily use.
- Radar scores are leads for a person to review, not predictions.
- Console is not on sale yet. If you want it for your own company, ask about early access.
Why it matters to you
The same habits go into every build.
Written tests, a person approving what matters, a trail you can read, and a cap on what the AI can spend. They are how we run our own company, and how we build yours.
More proof, with where it goes wrong: the proof notes.