AI memory and a context layer are not the same thing
Companies use the two terms interchangeably. Memory picks up what you tell an AI and stays behind in the tool that kept it. A context layer holds what your organization has delivered, and it works in the AI tools you already use. You only notice the difference when you switch tools or onboard a colleague.
Memory and context side by side
- IdmenMemory in the AI tool
- How it’s builtWe build it from all of your tools and systems: documents, email, conversations and workflows.It picks up what someone mentions in chats.
- Starting pointStarts full, with what already lives in your tools and systems.Starts empty and fills up over time.
- Where it livesSeparate from the tool. The layer sits underneath and connects through an open standard.Inside the platform that keeps it.
- Who controls the contentYou do. With permissions per client and per project.The platform decides what it remembers.
- When you switch toolsIt comes along. Same layer, different entry point.Starts from scratch.
- How it stays currentPim updates it, adds to it and removes what no longer applies.It only learns from new chats.
They don’t conflict. Memory remembers what you asked yesterday; the layer knows what your organization delivered in 2023, and who worked on it.
The gap
Where memory runs out
Every major AI platform has memory by now. ChatGPT remembers things from earlier chats, Claude builds up context within a project, Gemini picks up preferences. That saves time as long as you stay in that one tool. Ask what your organization has done before, and you run into three problems.
It starts empty
The first time, it knows nothing about you. After dozens of chats it knows your role, your clients and your writing style. That takes weeks, you can’t check where it got that knowledge, and the report a colleague delivered in 2023 isn’t in it.
It’s tied to one person
The platform decides what it keeps. In ChatGPT you can view and clean up that list, but you can’t take it to another provider. And the memory belongs to a user, not to an engagement, so you can’t enforce confidentiality per client.
It stays in one tool
What ChatGPT has built up about you stays in ChatGPT. Claude starts from zero next to it, because nothing outside those tools exists for both of them to draw on.
The alternative
The layer underneath
A context layer flips it around. One layer sits underneath all your tools, and every tool draws from it. We build it from every tool and system in your organization: documents, email, conversations, projects and workflows.
Every project, client and workflow gets its own page, and those pages link to each other. We set permissions per client and per project. Through an open standard, we connect that layer to ChatGPT or Claude, so the tool you already use backs up its answer with your organization’s work. The layer stays yours: you can export the source documents, structure, knowledge objects and audit log.
- One layer per organization
- Built from your own sources
- Under the AI you already use
- You stay the owner
At Idmen, that layer is called the Company Brain: one brain built from all of your tools and systems, getting smarter with every project. Pim keeps it up to date, and colleagues and agents draw on the same context, in whichever AI you use.
Explore the Company BrainHow it remembers
A brain stores facts, not a pile of documents
A document is what comes in: meeting notes, a report, an email. What the brain stores are the facts inside it. One set of meeting notes easily yields four, and each fact lands with the client, project or person it’s about.
Introductory meeting with Eva de Wit, CFO since this month at Rijnstroom. She wants a cost analysis of grid operations first before we send a proposal. We meet again on Tuesday, October 14. We agreed that the proposal goes past Joris first before it goes out the door.
What the brain knows about Rijnstroom
- Already thereCFO is Marc JanssenCFO is Eva de Wit
- AddedWants a cost analysis first
- Expires Oct 14Follow-up meeting on Tuesday
- NewProposals go past Joris first
Reading the notes
Update
When something changes, the new fact replaces the old one. The brain only keeps what’s true now, so no answer relies on a CFO who has already left.
Add
A new fact that adds something sits next to what was already there. With every set of notes, the client, the project and the people behind them gain depth.
Forget
A fact with a date expires once that date has passed. A follow-up meeting on Tuesday is no longer a meeting on Wednesday, so the brain removes it.
The switching problem
With memory in the tool
You work with ChatGPT for six months. It knows your role, your clients and your writing style. Then your organization moves to another tool, and that tool knows nothing at the first question. Memory belongs to the platform, so better onboarding won’t fix this. Every new tool starts empty, and so does every new colleague.
With a context layer underneath
The layer is separate from the tool. Connect a new tool and it knows what your organization knows from day one, with the same permissions and the same citations. If you stop working with us, you keep the layer; you lose the people who maintain it.
So the question is where your organization’s work lives, and whether every tool can reach it.
FAQ
Frequently asked questions
No. Memory grows out of past chats and stays in the platform that kept it. We build a context layer from your organization’s sources: structured, permissioned and yours. The two can coexist. Memory knows what you asked yesterday; the layer knows what your organization delivered.
Yes. They do different jobs. The memory in a tool tracks your chats inside that one tool. The context layer supplies what your organization knows before the first chat starts.
MCP (Model Context Protocol) is the open standard that lets an AI tool read from an external knowledge source. Connect the context layer to it, and ChatGPT or Claude pulls in what each question needs. Nobody copies and pastes, and you don’t add a second subscription.
Any tool that supports MCP, including ChatGPT and Claude. Pim also works with it in Teams and over email, where your team already discusses the work.
Give your AI the memory of your whole organization
In thirty minutes we show the brain answering live, on our demo environment or on questions from your own work.