Sharing context across your team’s AI assistants
Every member of your team has an AI assistant, and each one describes the company to it in their own way. Here is how to get all those AIs reading the same thing, without opening everything to everyone.
One team, five AIs, five versions of the company
In a team of five people who use an AI, there are currently five descriptions of the same client, five ways of explaining the product, and five ideas of what was decided last month. Each person wrote their own, one evening, in a conversation.
Nobody sees it, because every conversation is private. The effects, though, are visible.
- Two proposals sent to the same client don’t have the same tone, or quite the same prices.
- One person is still following the old procedure, because it is the only one their AI knows.
- A new starter takes three weeks to get usable answers from their assistant, the time it takes to rebuild what the others spent months teaching theirs.
- When someone leaves, everything they had explained to their AI leaves with them.
The problem is not the AI. It is that context, meaning what the assistant needs to know to answer well, has remained an individual matter while the work is collective.
What we mean by a team’s context
The word is vague. In practice, it covers five families of content.
- Identity. What the company does, who for, what it doesn’t do, its vocabulary, its tone.
- Methods. How you respond to a tender, how you prepare a workshop, how you handle a complaint.
- Files. What you know about a client, a project, a product: the brief, the meeting notes, the constraints.
- Decisions. What was settled, when, and why. This is the most neglected family and the most valuable.
- References. Templates, successful examples, prices, the regulations that apply.
All of this already exists in your organisation. Part of it is written down, scattered across shared folders. Another part is in people’s heads. And a third part, which is new, is in each person’s private conversations with their AI.
Why a shared folder is not enough
The first idea is always the same: “we already have a Drive, we just need to tell people to take the documents from there”. That doesn’t solve the problem, for four reasons.
The AI won’t go and look there by itself. Without a connection, it is still a person who opens the folder, finds the right file and pastes it into their conversation. The manual step is still there, and the omissions that come with it.
A pasted file is a copy. As soon as it is in a conversation or in an assistant’s project space, it no longer follows corrections. Three months later, out-of-date copies are living in a dozen places.
The assistant reads all or nothing. An attached file is swallowed whole, even if the question concerns a single paragraph. With long documents, that weighs on the quality of the answers and on usage.
The filing isn’t designed for an AI. “Meeting notes v3 final final” can just about be found when you know what you are looking for. An assistant needs clear titles, up-to-date documents and a way of knowing which one is authoritative.
A shared folder is still a good place to store files. It is not shared context.
The four conditions for truly shared context
A single source
Each piece of knowledge lives in one place, and only one. Not one copy per assistant, not one version per person. When it is corrected, it is corrected for everyone.
Direct reading by the assistants
Each member connects their AI to this source. The assistant does the looking, at the moment it needs to, without anyone pasting anything. The MCP protocol has made this possible with most assistants; we explain it in MCP explained simply.
Rights that follow the person
This is the point that makes people hesitate, and rightly so: if every AI reads the same base, does everyone see everything? No, provided that the assistant has exactly the rights of the person using it. The intern has no access to the salaries folder; neither does their AI. The consultant on assignment A cannot see assignment B; neither can their AI.
This means rights have to be checked on every request from the assistant, not once and for all at the moment of connection.
A record of what is read and written
When an AI can read and write in the shared base, you need to be able to answer two questions: who consulted what, and who changed what. An access log and a version history are enough, but they are essential.
The method for putting this in place
Step 1: choose a single subject to start with
Don’t launch a “knowledge management programme”. Take the subject the team re-explains to its AIs most often. In an agency, that is often the method for replying to clients. In a product team, it is the decisions already made. In a support team, the procedures.
Step 2: have the person who knows do the writing
The person who has mastered the subject writes it down, once. They can get their AI to help with the drafting, but they are the one who approves it. One document, one subject, a title that says what it contains. The writing method is in Building a knowledge base for AI.
Step 3: organise by project, and think about access from the start
Create one project for what is common to the whole team, and one project per client, per assignment or per product. The way you divide things into projects is also the way you divide access: you open a project to a person, not a file.
| Project | Who has access | What goes in it |
|---|---|---|
| House methods | The whole team | Identity, tone, procedures, templates |
| Client or assignment A | The assignment team, the client as a guest | Brief, meeting notes, decisions |
| Product | Product and support teams | Decisions, user feedback, roadmap |
| Management | Partners | Detailed prices, contracts, sensitive material |
Step 4: everyone connects their own assistant
Each member connects their AI with their own account. That is what guarantees that the assistant inherits the person’s rights, and that the log shows who did what. A shared account would ruin both.
Step 5: give the AIs a rule for writing
Reading carries no risk. Writing needs a rule. The simplest one: an AI can propose and record (a decision made in a meeting, a clarification obtained from a client), but it does not replace an authoritative document without the person having reviewed it. Write this rule in the shared project: the assistants will read it.
Step 6: build the habit
The rule that keeps the whole thing alive fits in one sentence: if you are about to explain it to your AI, write it in the project instead. The effort is the same. The difference is that next time, and for everyone else, it is already done.
The objections you always hear
“Our documents aren’t tidy enough.” They never will be until they are put to use. Start with one subject, write it well, and leave the rest where it is.
“People won’t write.” They already write, every day, in their private conversations. This is about moving that effort, not adding to it.
“We don’t all use the same AI.” That is exactly the case where a shared source that is independent of the assistants comes into its own. Everyone keeps their own tool.
“What about confidentiality?” It plays out at two levels. In the base: rights per project, checked on every request. With the assistant’s provider: what the AI reads is sent to them, under each person’s contract. Check the settings on your subscriptions, particularly on the use of data for training.
What this changes in practice
The most visible gain is time: the time you no longer spend pasting things in, and the time you no longer spend correcting an answer built on out-of-date information.
The second is consistency. When five assistants read the same method, five people produce work that looks alike.
The third is usage. An assistant that reads the relevant passage of a document uses less than an assistant that has the whole document pasted into it, in every conversation. We go into this in Reducing your token usage.
The last one shows up later: what the team knows no longer depends on any particular person being there.
How Linkevio meets these four conditions
Linkevio is built around this specific need.
- A single source. Your documents are written or imported into projects. A saved correction is the version that every assistant reads from then on. A document can be marked as authoritative, exploratory or superseded.
- Direct reading. Each member connects the assistant they already use. The AI searches, receives short excerpts with their location, and reads the relevant part.
- The person’s rights. The assistant has its user’s rights and not one more, and they are checked on every request. A client can be invited to a single project.
- A record. Every request is recorded in a log, documents keep their versions, and an assistant’s connection can be cut at any time.
The free plan lets two of you try it out on one project. Create your workspace, choose the subject your team re-explains most often, and run the test with a colleague. Details on data protection are on the Security and privacy page.
Frequently asked questions
Does the whole team have to use the same AI assistant?
No. The shared source is independent of the assistants. Each person connects the one they use, as long as it accepts connections using the MCP standard.
Can a colleague see my conversations with my AI?
No. What is shared is the documents in the projects, not the conversations. Your assistant reads the shared base; your exchanges with it stay with its provider, under your account.
How do you stop an AI writing any old thing in the base?
Through rights and through the history. A person who only has the right to read a project has an AI that can only read. And every change creates a version: you can see who changed what, and roll back if necessary.
What happens when someone leaves the team?
You remove their access: their assistant loses its access at the same instant. What they wrote in the projects stays with the team.
Can we invite a client?
Yes, to a specific project. They read and comment on what you open to them, without seeing the other projects.
Where do we start if there are only two or three of us?
With a single document: the method or description you have repeated most often. Write it, connect your assistants, and compare the answers. This is also a good time to read Why your AI forgets everything, which explains the underlying mechanism.