The Jarvis AI brain that remembers.

Recognise that log? Then you are probably in one of three situations. You work alone and explain to ChatGPT or Claude every morning what you do, what your rates are and how you sound. Or you work with a few colleagues and everyone has their own chats, so the same question gets three different answers. Or you work for several clients and knowledge must never bleed from one into the other. In all three cases the problem is the same: the knowledge sits in heads and separate windows, not in a place that people and AI both work from.

To begin you need very little. An account with at least one AI tool you already use, such as ChatGPT or Claude; that subscription stays yours and you pay for it with that provider. An email address. And for the free template, a program in which you can open a text file, such as Notepad on Windows or TextEdit on a Mac. More is not needed: the demo simply opens in your browser, without an account and without an install, and for the first two steps you need nothing technical.

The path is just as short. Step one costs no money and no account: download the company brain template and fill it in. Typing it yourself takes about an hour for a first version; let your AI ask you the questions and you are there in about ten minutes, and the ready made prompt for that is on the getting started page.

Step two is looking: open the interactive demo and see what the same idea looks like as a system. Only after that does a draft plan come into view, with Brain Start currently shown at 9 euro per month before any country-specific tax treatment. Country-specific tax treatment remains pending. Online payment and ordering remain closed. You can only join the non-binding waiting list with a preferred plan. That is not an order and creates no agreement or payment obligation.

An organisation brain is the layer between your company and the AI tools you use. Not a chatbot, and not a new model of ours. Think of it as the place that holds what applies inside your company: how you work, what you agree, which decisions have been taken, which sources belong to them and who approved something. The moment an AI starts work, it collects that context there instead of asking you for it again. Hence the other name we use for it, AI company brain: the brain belongs to the company, the AI tools are the hands.

The difference with a folder full of documents comes down to four things. First, every piece of knowledge has a status: proposal, decision or lasting knowledge. Second, every piece of knowledge has a source, an owner, a validity and a version, so you can check where something came from. Third, knowledge is bounded: an item belongs to a scope and is visible only to the people and agents it concerns. And fourth, the brain is readable not only by people but by the AI tools connected to it, in a form they can actually use.

That also says what it is not. Neither a note taking app nor a personal second brain in the sense the well known note tools mean it: those are built for one head, and the whole point here is that knowledge is shared. Nor is it a document search engine that combs your files and pulls an answer out of them. A knowledge base like that answers questions about documents. An organisation brain keeps what was decided, who decided it and what has happened to it since, and drives the work with that. Those are two different problems, and the second one survives after you solve the first.

The reason this becomes urgent now is the sum of the parts. One AI tool with its own memory is manageable. With three tools, five colleagues and ten running projects it stops being clear which version of the truth is floating around where. The memory in ChatGPT lives in one tool and one account. What you record there, Claude does not know, and what a colleague records, your account does not know. Jarvis solves that by taking the memory away from the tools and giving it to the organisation, after which every tool is welcome to use it again.

You can try this without software first. The free company brain template is the manual version: one fillable file you hand to Claude, ChatGPT, Codex or Cursor, so your AI at least knows who you are and how you work. That works fine until the moment several people have to edit it, versions appear side by side and nobody knows which file is the real one. And that moment is where Jarvis starts. Not sure yet whether you need this? Work through the decision guide in ChatGPT memory or your own AI brain (in Dutch) first: five questions that show whether the built in memory of your tools is enough for you.

Anyone working with AI arrives at skills sooner or later. A skill is an instruction file that gives an AI assistant a fixed way of working for one task: how you build a quote, how you run an analysis, what it must never invent. We give a hundred of them away in the skill library, and they are worth having. But a skill solves only half the problem.

A skill teaches your AI how it does something. It tells it nothing about your company, and it remembers nothing. Every new session starts empty again: you supply the context again, you explain your rates again, and you check again that it has not invented anything. So the skill is neither the engine nor the memory. The brain is where the memory lives.

That difference is easiest to see through one and the same question, asked two days apart.

So the two layers work together instead of against each other. The skill keeps doing what it is good at: it gives the AI a method and a list of things it refuses. The brain supplies the context that makes the method worth anything. Run a quote skill without a brain and you fill in your rates, your tone and your terms every time. Run it with a brain and they are already there, in the version that applies right now. Nothing about the skill itself changes; only the amount of typing does.

There is a third step. If you want this work ready every day without having to sit in front of a chat window yourself, that is a service rather than a product: the AI employee, from €950 per month, which we deliver through Mansotti, the business of which TheSEO is the trading name. The same rule holds there as in Jarvis: output stays a draft until a person signs it off. Jarvis is the layer underneath. The cheapest step, and at the same time the one that makes the other two better.

The most expensive mistake you can make with AI is not a wrong answer. What costs you is a wrong answer that looks certain and then gets copied by everyone. In a shared brain that risk is bigger than in a single chat window, because whatever ends up in there is read by everyone and by every connected AI. Which is why provenance is not an extra field here but a condition.

In practice that means three things. Every piece of knowledge carries its source, owner, scope, validity and version with it. When sources contradict each other, Jarvis does not quietly store one answer as the truth but asks for a judgement. And an AI may make a proposal, but the step from proposal to lasting knowledge always runs past a person. The deep dive on provenance writes out how those fields relate to each other.

The other half of the story is version control. Knowledge in a business changes constantly: rates, terms of delivery, who handles which client, what you will and will not take on. If a change overwrites the previous version, you lose exactly the information you need at the moment something goes wrong. So Jarvis keeps the old version next to the new one, with a date and an approver, so that afterwards you can see which agreement applied when someone acted.

Not everything we write here has been demonstrably tested in production. Some parts are product principles that sit in the interface, others are still release gates: negative cross tenant tests, an external security review, a restore and delete exercise. On jarvis/veiligheid each point says whether it is policy, interface or still to be proven, and the trust centre holds the documents. Documents that are still drafts carry the word draft in their heading, so you can tell from the document itself where you stand.

A great deal is being promised around AI at the moment, and most of those promises cannot be checked. So here is a list of what we do say and what we deliberately do not say. You may hold us to the first list.

This is what you may expect from us. The pricing model is a checkable property: there is no meter on prompts, tokens or model calls, and your own AI subscriptions stay yours. Human approval, provenance, version control and an isolated brain per organisation are product functions and not conditions in small print. The interactive demo is the real interface with fictional sample data and works without an account. And we run on it ourselves every day, with every session, task and decision logged and readable afterwards.

This is what you will not hear from us. No client count, because that is not an honest selling point right now. No percentages about time saved or revenue growth, because we have not measured them. No timescales within which something works. No certifications, and no standard belonging to someone else that we put forward instead: naming a standard is not a certificate and it proves nothing about us. As long as there is no valid external certification with a clear scope, we will not sell Jarvis as such either.

And this is what is not there yet. Automatic checkout is deliberately not released. Paying is the last step with us and not the first: that button opens once contract, tax, provisioning and revocation all follow the same subscription status. The prices themselves also carry the status proposition in validation at the moment, which means the amounts named are the current plans and have not been fixed for good. Beyond that a few things are still open which are named in full on the security page, among them the negative cross tenant test set and an external security review.

You can hold us to those three lists. If the status of a point changes, it changes with it in the change register, with a date and a reason. How we draw this line across the whole site is in the knowledge base and on about us.

The problem is not that your AI is too stupid, but that it remembers nothing about your company. You solve that problem with memory, not with a better model or a longer prompt.

The cheapest way to notice that is the free company brain template (in Dutch). Fill it in, hand it to your AI and see what changes in the first session. If that works, and you then run into versions, colleagues and several tools, the demo is the next step. The non-binding waiting list can then record your preferred trial plan. Joining is not an order and creates no agreement or payment obligation. Would you rather speak to someone first about what AI could do in your situation? That works too.