Business is not released yet. Business testing opens in December 2026, and launch is planned for the first quarter of 2027, with January the goal. Book a consultation to see where it is now and to shape what ships.
Business Overview
Illustration of planned workAn illustration with made-up figures, showing planned work. The app does not connect to Xero, Gmail, timesheets or a bank today.
The one you brief. Everything else answers to it. You ask once. It works out the rest.
Next: the whole thing drawn out, end to end.
Cortex is the one you brief, and the only one you have to. Every agent in your business reports to it, the way a department head reports to a manager. You do not pick which agent to talk to, learn what each one is called, or chase any of them for an answer, because that is its job rather than yours.
Give it a sentence and it works out what the job actually needs: which figures matter, who holds them, what order the work goes in, and what it has to go and ask for. It comes back with one answer rather than five loose ends, and anything that needs a person is held for you rather than done anyway.
It cannot see anything it has not been given. Cortex works from the notes and files in your vault, and nothing else. Access that differs by person or by department is planned for Business and is not built yet.
The same job, drawn from the moment you ask to the moment you get an answer back.
Illustration of planned work: department access, role-scoped agents and team messaging are not in the app today.
Decides what the job needs, who does which part, and what it has to go and ask for.
the gateCortex can only read a department's vault if that department grants access.
Checked on every request, not once at setup. A refusal is recorded like anything else.Supplier invoices, last two quarters. Nothing else.
Ad spend by channel, same period. Nothing else.
Nothing read yet. Cortex works around it or waits, and tells you which.
and the departments message each other directly, without going back through you (planned)
This one gets the job done. Everything after the answer is where the work actually happens: doing it, not describing it.
The difference shows up on the second question, not the first. A chat window starts from nothing every time; this starts from everything it has already been told, so you stop re-explaining your own business to it every morning.
An agent is a worker made of software. Each one runs inside permissions you set: it reads freely and acts only where you have allowed it.
Permissions are set per action, not per person, so "read the sales figures" and "email the client" are separate decisions even for the same agent. Reading is usually open; anything that spends money, changes a price or leaves the building waits for a named person.
A full trail of what each agent touched, what it changed and when, kept on hardware you own.
Files are snapshotted before an agent writes to them, so a bad run rewinds byte for byte rather than being argued about afterwards. The record is the point: an assistant nobody can audit is one nobody can let near the parts of a business that matter.
First, what it is. NeuralVault Business is an application you own and run, not a service we operate for you. It installs on hardware you control and your data stays there. It is the same application as Personal, with the layer a company needs built on top.
It surfaces costs that have been quietly adding up. It connects things that sit in different systems and never get compared. And it answers questions about your business specifically, because what it knows is your operation, not the internet.
This is planned for Business, and it is not built yet. The idea is that nothing about the arrangement changes when you hire. Each person would join at the access level you give them, and nobody would have to learn a new way of asking for things. The app today is built for one person on one machine.
Planned: one person or a full team, everyone talks to the same Cortex, which directs every specialist agent. The access levels shown are the plan for Business, not a feature of the app today.
NeuralVault runs on the computers your team already uses. A model is the AI itself, the part that does the actual thinking. A desktop has room to run more of them on your own machine, which gives NeuralVault more it can do and more freedom to get on with it. It is an upgrade, not a requirement.
Nothing to install in a server room, and nothing you are locked out of.
Room to run more of the AI on your own machine, so it can take on more and simply get on with it.
A desktop is a recommendation, not a requirement. Either way, your data, your dashboard and the task board the agents work from all sit on hardware you own, and physical access stays with you.
Reading is usually open, because an agent that cannot see your numbers cannot help you with them. Anything that changes a number, spends money, or leaves your business stops and waits for a person.
Permissions
Early previewAgents cannot mark their own work as done. Finished tasks sit in Awaiting Review until a person accepts them.
Move an action between allowed and ask-first whenever you like. Nothing needs rebuilding for a permission to change.
When an agent reaches a boundary it stops and asks. It does not quietly pick the safest-looking option on your behalf and tell you afterwards.
What was touched, what changed, and when. Not a summary after the fact, a record kept as the work happens.
Shopify, YouTube, Meta Ads, Search Console, GitHub and RSS, coming soon: NeuralVault will read the systems you already sell and publish through, and put revenue, ad spend, channel stats and tracked costs on a single page that updates itself. No more opening six tabs to know how the week went.
The chart below is an illustration of that page, drawn from generated sample data and not from any real system. Pick a metric and it redraws around it. Pick a timeframe and it redraws again: hour by hour when something is going wrong today, week by week for how the quarter is tracking, year by year when you are planning ahead. The totals recalculate with it, so the number you are reading always belongs to the window you are looking at.
The design is that nothing on the page is typed in or exported: once the connectors ship, every point comes from a system you already use, so it is current when you open it rather than as current as the last time somebody built a report. Today the page ships and every figure reads empty.
It is not one assistant answering questions. It is a set of them that can reach the systems you already run, remember what happened last week, act only where you have allowed it, and leave a record of everything they did. The plan for Business is that every employee gets the same coordinated team of AI agents, working through your tasks while you stay in control. Per-user accounts are not built yet. Every capability below runs on hardware you own.
Assistant
Agent team
Pipeline
Approvals
Comms
Privacy
Every invoice, every message, every decision and every note your agents have ever written is held in one place and indexed. So "what actually happened with that supplier last March" stops being an afternoon of digging through folders and becomes a question you ask in passing.
The vault is indexed once, and after that the index keeps itself current as things change. Nothing is uploaded to do it: the index sits beside your data, on your hardware.
Most teams have already answered the AI question without being asked: staff paste client data into public chatbots because nobody gave them a sanctioned tool that does the same job safely. Deployment is how you replace that with something that stays inside your walls. The software does not care where you are, so this part is not limited to Essex.
Most AI programmes do not fail because the model was not good enough. They fail because the tool arrived without anybody being shown how to use it, and because the only thing measured afterwards was how much of it got used. That is a management problem wearing a technology costume, and buying a better model does not touch it.
Several large firms ranked staff by how much AI they consumed. Work appeared that existed only to move the number. The rankings are gone. The bills were not.
An agent handed a vague instruction does not stop and ask. It loops, correcting an instruction that was wrong to begin with. That is a training gap, and no amount of licence spend closes it.
The fastest way to kill honest AI use is to pile the freed hours straight back on. People notice, and they quietly stop telling you what they automated.
So a deployment here is scoped with you, installed on your hardware, and handed over with your team actually shown how to use it, which is step four above rather than an afterthought. And the thing worth watching afterwards is what got produced, not how many tokens got spent. The consultation is where we work out which of these you already have, and it costs nothing.
AI Proficiency Training is for the people who will actually use it, not for the person who signed it off. A system your staff log into and work through at their own pace: a video and a written explanation of each concept, then tasks where they have to solve a real problem using an AI and show what they got. Nobody passes by watching. And it is taught by the people who deployed the system, on your own work, rather than by a training firm that has never seen your business. The band above is why it exists. Rollouts fail on this, and it is the half nobody sells.
How to brief a tool so it does the right thing first time, and how to spot an answer that is confidently wrong before it reaches a client.
The part the big firms are spending on: getting your own methods into it. Writing the instruction file, and recording a job once so the agents have the steps rather than a generic answer.
How to give it your documents and records so it knows the business, and what it genuinely can and cannot learn from them.
Where the cost actually goes, and how to keep the bill tracking the work rather than the enthusiasm. This is the failure the section above describes.
The board, the agents, the approval cards and the record, taught as the job somebody does on a Tuesday rather than as a tour of the menus.
The tasks are the assessment. A qualification per person, earned on work they actually completed, so you can see who is ready to be trusted with it rather than who sat through the session.
It is under development and it is not part of the current offer. There is no date on it, nothing in a deployment today depends on it, and you are not being asked to wait for it. When it does land it will be priced per seat, or per department where a whole team goes through it, and it splits: a cheap general course everybody sits, a dearer specialist course built on your own systems that you would put a handful of people through, and a separately charged look at those systems first. It will be available on its own as well as alongside a deployment. There is no figure to quote yet and we would rather say so than invent one. What it would cover, in full, or say at the consultation if it would matter to you, because that is the sort of thing that decides what gets finished first. Arrange the consultation.
Two things get priced: the application, and the deployment around it. The app is the product, and it is yours. Hardware, setup, migration and ongoing support are an optional layer on top, and how much of that you actually need is why every deployment is scoped and quoted after the free consultation.
You get a fixed number before any work starts, and if we do not believe the deployment will pay for itself, we tell you then.
Ready to talk, or want to see it working first? That conversation costs nothing and there is no obligation at the end of it.
Not certain Business is the one you want? See all three editions side by side, Free included.
We will email you when there is something worth knowing.Major updates and the day Business is ready to deploy. No newsletter, no drip sequence, and your address goes nowhere else.
Thanks. You will hear from us when there is real news.
Takes about four seconds. No account, no card.
No spam, no drip sequence. Pick what you want to hear about and nothing else.