How it works — explained in plain English
Nothing to install.
Nothing you need to know how to do.
You connect a memory.
This page explains — without IT jargon — what exactly gets connected to ChatGPT or Claude, what data is sent and where, where the extra knowledge comes from and who makes sure it stays fresh. In a way that a teacher, a doctor, a solicitor and the man from the garage can all follow. Because they are exactly who we built it for.
open round the clock
The whole "connection" is typing in one address and a password. Once.
Step 1 — what you are actually connecting
The chat you use gets access to your private base
ChatGPT and Claude are like an exceptionally well-read colleague: they have read half the world, but they do not know your business. They do not know what is in your contracts, price lists and notes — because they have never seen them.
Memory is your own knowledge base — think of it as a well-organised ring binder that sits on a server and is open round the clock, whether or not your computer is switched on. That binder is what the chat gets access to.
Connecting is simpler than it sounds: in your chat's settings you enter one address (we give it to you when you buy) and log in with a password. You do not install any software and you change nothing on your computer. The connection is saved to your chat account, so it then works wherever you log in — on your laptop and on your phone.
Chats can do this because a shared, open standard for such connections now exists — like one plug shape that fits many devices. That standard is called MCP. Hence our name: WiseMCP.
Step 2 — where the base gets its knowledge
Three streams of knowledge. Only one needs work from you.
The base does not fill itself out of thin air — knowledge flows into it by three routes. Only the first is yours; the assistant looks after the other two, like a meticulous secretary who never forgets anything.
You upload documents
Through an ordinary browser, just as you would attach a file to an email. PDFs, Word documents, scans — the system turns them into text that can then be searched. Even when a document runs to several hundred pages.
Conversation notes
When you settle something important while talking to the chat — a rate, a decision, a deadline — the assistant saves it as a note by itself. Before saving, it checks whether a similar note already exists, to avoid clutter.
Web pages
Seen an article online that is worth keeping? You give the address and the system fetches the page content and stores it in the base — the original, not a summary, so nothing gets distorted.
All entries link to one another like entries in an encyclopaedia: from a note about a customer you can move to their contract and to the latest agreements. So when the assistant answers a question, it sees not a single sheet of paper but the whole connected story.
Step 3 — what data is sent
The journey of one question, from keyboard to answer
You type a question in the chat
For example: "What rate did we agree with the customer from Leeds?" The conversation takes place where it always does — in ChatGPT or Claude.
The chat asks your base
Over an encrypted connection — the same kind online banking uses — the chat passes the question to your base. Only the chat has access, because only it has logged in with your password.
The base sends back matching passages
The base does not send the whole archive — it picks the few passages from documents and notes that best match the question, and sends back only those. A whole document travels only when the assistant asks for it explicitly.
The chat composes the answer and cites its sources
From the passages it receives, the assistant builds an answer — and it can point to the document or note each piece of information came from. You do not have to take its word for it; you can check.
So exactly two things travel between the chat and your base: the text of the question one way and matching passages of your knowledge the other. Always over an encrypted connection, always behind a password. Your base is separate — every company gets its own, which nobody else can enter.
To be completely clear: the chat conversation itself works as it always has — it is handled by OpenAI (ChatGPT) or Anthropic (Claude), on their terms, which you accept by using the chat. Memory adds your knowledge to that conversation — and looks after it on its own side.
Step 4 — federated bases
Branches of extra knowledge spread out from your main base
Your main base knows your business. But there is a great deal you do not have to collect yourself — trade, how-to and subject knowledge. That is why we run subject bases: separate collections of knowledge, each on one topic, which we fill and keep up to date.
The best way to picture it: your base is your bookcase at home, and the subject bases are shelves in a reading room. In the panel you use a switch to mark which shelves you want to use — usually one to five. From then on, when your assistant looks for an answer, it goes through your bookcase and the chosen shelves at the same time, and marks where each piece of information came from.
your company only
One question searches the main base and every enabled shelf at once.
Why is this clever? Because one shelf serves all customers at once. We do not copy it into each company separately — so when we update it, everyone benefits from the fresh version straight away, and the cost is spread across many users.
And privacy? Since many companies use the same shelf, your question should not reach it in readable form — and it does not. Only a numerical fingerprint of the question is sent to a shared shelf: a string of numbers that makes it possible to find matching content but is not the text of your question. The readable text of the question does not leave your base — that is how the system is built; it is not a promise, it is the design.
Billing is fair: shelves have no separate subscription. Using them draws on the same paid units as ordinary questions — and the panel shows in black and white how much each enabled shelf has used. No longer worth it? You flip the switch off.
Step 5 — agents
Short programs that work behind the scenes
In the world of computers an "agent" is a short program that has been given one task and carries it out on its own — like an employee with a single job, only tireless. In our system agents work in two places.
On our side — at the subject shelves
Agents search for content on a given topic, put it in order and prepare it to be added to the right shelf. Thanks to them the shelves grow and do not go out of date. A person keeps watch over what finally goes on the shelf — the agent is an assistant, not the editor-in-chief.
On your side — in your base
The assistant itself plays the part of the everyday agent: it takes notes from conversations, checks for duplicates, adds links between entries and, when you ask, pulls pages from the internet. And on request, additional agents can be set up in your base to carry out specific work on your knowledge.
The closest real-life comparison: librarians who, after the reading room closes, put the books back on the right shelves and add index cards. In the morning everything is in its place — although nobody saw when it happened.
The rules — what this system does not do
Four lines we do not cross
It does not change your documents
The files you have uploaded are read-only for the assistant. It cannot correct, overwrite or delete them — that limit is built into the system, not written into the terms of use. The assistant only changes notes it has created itself.
It does not show them to anyone else
Every company has its own isolated base under its own address and password. Your documents do not go onto any shared shelf and are not visible to other customers.
It does not send your questions to shared shelves
Only the numerical fingerprint of a question goes out to the subject bases that many companies use — never its text in readable form.
It does not add charges quietly
You pay for measured usage, not for estimates. The panel shows a day-by-day chart and the share of each enabled shelf. No hidden items.
Finally — getting started
Three steps and the chat knows your business
You receive an address and a password
When you buy, you get the address of your Memory base and login details for the panel. The base starts straight away and runs round the clock.
You enter the address in your chat's settings
In ChatGPT or Claude you add a connection, paste the address and log in with your password. You do this once — after that it works on every one of your devices.
You upload documents and talk
You upload files through your browser. In the panel you switch on the subject shelves that suit your trade. From then on you simply ask — and the chat answers from your knowledge.
Connect your knowledge to the conversation
WiseMCP — Memory. Your knowledge base and lasting memory for ChatGPT and Claude. One address, one password — and a chat that finally knows your business.
Model Context Protocol – MCP
Install the ChatGPT desktop app
Download the desktop version of ChatGPT for Mac or Windows from the OpenAI website.
Connect your WiseMCP service
Sign up for our service and connect your business memory to the ChatGPT app you have installed.
Work in ChatGPT with your own business data
Add your PDF and DOCX files, presentations and everything else to the memory.