Jarvis: the AI brain for ChatGPT and Claude

Jarvis is not a new AI tool. It is a memory layer that sits between ChatGPT and Claude. Rather than letting your knowledge scatter across separate chats, Jarvis stores your company knowledge, projects, tasks and decisions centrally in a database.

The connection runs over the Model Context Protocol (MCP). That was a deliberate design choice: ChatGPT and Claude connect to the same database, so both tools work from exactly the same context. Note where things stand further down: that connection to your own brain is still behind a release gate and has not been opened up. Start a new session and the AI collects the relevant background by itself, rather than you explaining everything again.

That makes Jarvis valuable above all if you combine improving your SEO and other long running programmes with AI. Context about strategy, clients and content then stays put, even when you switch between tools.

The first phase of AI was about prompts. The second phase is about context. Anyone can open a chat window and ask a question; that is no longer what makes the difference. The difference appears the moment an AI genuinely knows your situation. Not just your question today, but the decisions from last week, the projects still open, the clients you are working with, your writing style and the way you normally make choices.

At most companies this is where it goes wrong. The strategy is in ChatGPT. The technical work is in Claude. The tasks are in a document. The last decision is in a chat from three weeks ago. Want to pick it up a day later and you spend ten minutes hunting down context before you get to the real work. That failure is not a model problem, it is a memory problem.

An AI second brain solves that by taking the context out of the separate windows and putting it in one place. Four things make such a brain usable rather than merely tidy.

• Continuity. More tools, longer prompts or another dashboard do not solve context loss. One fixed place where the context stays does.
• Context over search. Without context you get general answers. With context you get advice that fits your offer, your clients and your earlier choices.
• Separate chats are not a system. For one task a chat is fine. As soon as you have several projects, noise appears: you lose the overview, you repeat yourself and you build nothing lasting.
• It gets better the more you use it. A prompt stays the same. A brain grows: every decision, every correction and every project update sharpens it.

Which is the difference between a chatbot and an AI second brain. The chatbot answers the question you are asking now. The brain also knows where you came from and what you are working on.

A prompt tells an AI what to do now. A prompt bundle is a collection of those: a hundred good instructions you can copy. Useful, and we give a hundred of them away in the skill library. But a prompt knows nothing about your company and remembers nothing from yesterday.

Jarvis works a layer deeper. Instead of instructions, it keeps the structure those instructions draw on: your fixed core, your projects, your tasks and the decisions you have taken, each with a source and a version. 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.

In practice that means: you keep using the same skills and the same prompts. What changes is that you no longer have to supply the context with them. For the difference between those two layers in more detail, you will find it fully worked out on the how it works page.

It was built for anyone who notices that AI is powerful but that the workflow around it keeps fragmenting. Concretely that means:

• Business owners with several projects or companies, who want an overview of opportunities, clients and open actions.
• Marketers who have to manage SEO strategy and content clusters without starting over each time.
• Developers who want to share project memory between different AI tools.
• Consultants who want client context and earlier advice at hand.
• Makers who want to keep fixed formats and script styles.
• Teams who want to share the same context together, rather than working separately.

It looks different per role. This is what people put in it in practice.

• Business owners: company context, new opportunities, open actions, morning briefings and decisions taken.
• Marketers: SEO strategy, content clusters, campaign context you reuse, and the brand style nothing may deviate from.
• Developers: architecture choices, bug context, technical decisions and the next steps already waiting for the next session.
• Consultants: client context, advice already given, meeting notes and follow-ups that do not vanish into an old chat.
• Makers: fixed formats, script styles, ideas you pick up later and the creative direction you want to protect.
• Teams: shared context, tasks in one place, less handover and more consistency between people who do not speak daily.

The common thread: as soon as you work with AI on something that lasts longer than one conversation, a central memory becomes useful.

ChatGPT and Claude both have a memory feature of their own, and those have got a great deal better in recent years. The question is not whether they work, but where they stop. Four common questions, with the answer attached.

Why the memory in ChatGPT is not enough

The memory in ChatGPT is tied to ChatGPT, and inside ChatGPT to your account. What you record there, Claude does not know, and what your colleague records, your account does not know. Add a coding tool, a task list or a dashboard to that and fragmentation appears anyway, only at a higher level.

Why the memory in Claude is not enough

Claude can handle a lot of context inside a project and does it well. But your company does not live only in Claude. Most people use different AI tools for different tasks, and then you need a layer that sits above those tools rather than inside one of them.

And what about separate notes?

A note taking app stores information but does not activate it. The note only comes back when you look it up yourself, and that is what you forget on the busy days. A brain is built precisely to bring knowledge back at the right moment, in the session where it is needed.

And a dashboard on its own?

A dashboard is only worth something once it is connected to the real work. The screen is the cockpit; the value sits in the memory layer underneath. A cockpit without a memory is a fine overview of nothing.

Exactly what the built in memory features do and do not cover, and at what point an external brain begins, is worked out in AI memory between chats (in Dutch). Are you unsure whether you need this yet, the five decision questions in ChatGPT memory or your own AI brain (in Dutch) walk you through it.

Jarvis turns on five principles that together form the core of the system.

• Continuity. One fixed place where context stays put solves the fragmentation between separate chats.
• Context as the engine. An AI that knows who you are, who your clients are and how you work gives better advice than an AI that starts from zero.
• Separate chats are not a system. Structure only appears once projects are kept centrally.
• Bringing known information back. The AI remembers your offer, your clients and your open actions, instead of you repeating everything.
• A growing system. Every correction, task and update sharpens the memory.

Building a working system runs in six steps.

• Record the fixed core: who you are, which companies you have, your audience, your offer, your tone of voice and your rules.
• Order your projects: per project the goal, the status, open actions and decisions taken.
• Make tasks visible: blockers and actions stay findable instead of vanishing into a chat history.
• Start the AI with a briefing: at the start of a session the AI collects the right context by itself.
• Connect ChatGPT and Claude: a shared memory layer stops you doing the same work twice.
• Let the brain grow: every correction and update makes the system better over time.

Under the bonnet Jarvis is not a chatbot but a memory layer between you and your AI. Six parts together make the difference with a folder full of documents.

• A layer on top of your AI tools. You keep working in ChatGPT and Claude. The difference is that both draw from the same layer, rather than each tool keeping its own separate conversations.
• Context is stored, not typed again. Your core knowledge, projects, tasks and decisions sit in one place. What you record once stays available for every session.
• Recall at the start of a session. When a conversation starts, the connected AI collects what is relevant for that project, instead of you pasting it again.
• Writing back at the end. What was decided and done goes back into the memory, so the next session knows it.
• A boundary of its own per organisation. Every organisation brain has its own members, sources and history. Knowledge does not flow between them by itself; how that is set up is in super brain and client brains.
• A person between proposal and truth. An AI may make proposals, but the step from proposal to lasting knowledge runs past an authorised person. That is a technique and not a promise; it is worked out in human approval.

The part people underestimate most often is the fifth. As long as knowledge carries no owner, version and validity, nobody knows whether it still holds and who was allowed to change it. Why that is a condition rather than an extra field is in provenance.

Not every situation calls for a central memory. This is how you weigh it up:

• A one-off question: a single chat is usually good enough, a note taking app is not needed.
• A project lasting several weeks: separate chats get messy, a note taking app only stores information, Jarvis keeps the context active.
• Using ChatGPT and Claude together: without Jarvis there is no shared memory and you have to copy by hand, with Jarvis both tools work from the same memory layer.
• Not forgetting open tasks: in a single chat they disappear, in a note taking app they stay visible only if you keep them up yourself, in Jarvis they stay in view.
• AI as a full assistant: a single chat is too reactive, Jarvis supplies context, a briefing and direction.

You do not have to invest anything to notice what this does. The manual version is free, stays yours and works today. Five steps, and you are there in about an hour, or in ten minutes if you let your AI ask you the questions.

• Download the company brain template. One fillable markdown file, without an account and without an email address.
• Put it into your AI. In Claude as a skill or as project knowledge, in ChatGPT in a project or a custom GPT, in Codex and Cursor as a context file.
• Fill in your fixed core. Who you are, what you do, how you sound, what you never say, what your services cost and who decides what.
• Just carry on working. Every time something changes, you update the file and hand over the new version.
• Then see how it works as a system. The interactive demo is the real product interface with fictional sample data and opens without an account.

Where this seizes up is predictable and it is not the file's fault: you are the version control. With three tools and two colleagues you get copies that drift apart, and the file never warns you that it has gone stale. That is exactly where the paid product starts. The four plans begin at Brain Start for €9 per month including VAT; the full table is on Jarvis pricing and carries the status proposition in validation there.

The Model Context Protocol, MCP for short, is the standard by which an AI assistant talks to an outside source directly rather than through copying and pasting. This technique is what the connection between your AI and your brain rests on. Here the difference between what runs today and what still sits behind a release gate matters, because it saves you a disappointment.

What you can already connect today

Our public MCP server is running and is open source. Add it once to Claude, Claude Code or another MCP client and from that moment your AI can run an SEO scan itself, look up free skills and fetch the company brain template. No account and no key. There is a local findability check in there too, but it currently returns an error: the connection with the Google business register is not switched on, and the same holds for the same check on the site.

Five of the seven tools send nothing on; the SEO scan and the local findability check pass your input on to our site, which records the scanned address and the outcome, exactly as the tool on theseo.nl does. The endpoint, the instructions and that explanation are on the MCP page.

What is still behind a release gate

The personal memory connection, where ChatGPT and Claude read and update your own organisation brain, is designed but has not been opened up. Authentication, live connectors, tenant provisioning and payment sit in the same set of gates, and all of them have to be green first. As long as they are not, we hand out no keys for it and we promise no date with it. Which gates those are and what is still missing per gate is named in full on the security page and in the machine readable status in the documentation.

What you can look at in full in the meantime is the working interface: the interactive demo runs on fictional sample data and shows jobs, approvals, memory proposals and knowledge reviews. The screens are walked through one by one in the platform tour.

You do not have to understand this to work with it. But for anyone who wants to know: these are the six parts an organisation brain is built from, and why all six are in there.

• A central memory layer. Projects, core knowledge, tasks, decisions and messages sit in a database instead of in separate chats. That is the condition for everything below.
• Looking things up by keyword and by meaning at once. You find a decision even when you word it slightly differently from the person who wrote it down.
• Recall at a new conversation. The connected AI collects what is relevant for that project, not everything the brain knows. Scope is the brake here, not a limitation.
• Logging on close. What was decided and done the AI writes back, with the source and the time, so the trail stays followable.
• Walled off per organisation. Every organisation has its own boundary with its own members, sources and history. What you learn for one does not leak into the work for the other.
• It grows with you. Every choice and update sharpens the memory, provided someone keeps it up. A brain nobody maintains ages just as fast as a file.

The last point is the most honest warning we can give. This is not a system that fills itself. It is a system that remembers what you put into it, and that is exactly where a chat window falls short.

Jarvis is meant for people who notice that AI keeps getting more important while their way of working stays fragmented. You use ChatGPT for ideas, Claude for the work itself, maybe a coding tool alongside, a dashboard for the overview and separate documents for notes. That works fine while everything stays small. The moment you have several clients, projects or companies, the scattered setup becomes the bottleneck.

An AI second brain brings structure to that. It makes sure your company does not sit only in your head but is also available to the tools you use. That way you switch faster, you look things up less, you ask better questions and the AI can judge better which answer fits your situation.

The main benefit is not that information gets stored. The point is that the information becomes usable. An old note is only worth something once it comes back at the right moment. A decision is only useful once your AI knows it has already been taken. A task is only genuinely visible once it does not disappear into an old chat. That is why an organisation brain turns on memory, tasks, briefings and project status together, and not on one of those four separately.

ChatGPT and Claude keep doing what they do well throughout. What they gain is continuity. For business owners that can be the difference between AI as occasional help and AI as a daily working partner. The free company brain template is the first step to building that, and Jarvis is the step where you no longer have to keep it up by hand.

Three routes, and the first costs nothing. Start with the free company brain template and hand it to your AI: you notice the difference in the first session. If you want to see what the same idea looks like as a system, open the interactive demo, without an account. If you are ready for a brain you do not have to keep up yourself, the four plans are on Jarvis pricing and getting started explains how you order.

If you would first like to know whether this fits your situation, we will walk through it together. Get in touch or book a call. If it is about business arrangements, several organisation brains or bespoke work, we go through usage, plan, sources and sign-off in that conversation.