Teams & Organizations
Why personal memory must not automatically become company memory
Short answer
A personal memory holds drafts, doubts and false starts that were never meant to become the company’s voice. Merging it automatically costs you the candor of the people writing in it and blurs where things came from. The safer principle: what starts private does not leave private on its own, and what goes to the shared side goes through an explicit, filtered step.

You use an AI to prepare meetings, rough out drafts, think out loud. Your company, meanwhile, would love for all that work to benefit everyone. The idea comes quickly: plug every personal memory into one big shared memory, and continuity is taken care of.
It is not that simple, and the risk is not where people expect it. Our position fits in one sentence: what starts private does not leave private on its own. Sharing is possible, and often a good idea, as long as it is a deliberate step and not a leak. Here is why, what an organization can reasonably expect, and how to handle the move from personal to shared.
Why merging everything is tempting
The argument is a serious one. When someone changes roles or goes on leave, part of a project’s history leaves with them: what they decided, why, what they ruled out. More and more, that history sits in their conversations with an AI. Pouring it all into a common pool feels like rescuing everything at once, at no cost to anybody.
It is also the path of least resistance for a tool: sweeping up everything that passes through is easier than asking each person to sort. But now a working notebook has quietly become a collective archive, and the person who wrote it never decided that. Nobody said yes, and yet everyone can read.
What gets lost when private becomes shared
Candor goes first. A space you know others will read is no longer a space where you think out loud. You stop telling your AI that you are unsure, that you went down the wrong track, that an instruction from management looks absurd. You write for an audience, and the best questions are often the ones you would never ask in front of one.
Next goes the right to be wrong. A draft is supposed to be wrong. In a shared memory it can pass for the company’s position, simply because it is written down, dated and easy to find.
Then there are two more concrete risks. The first is leaking private information: a medical appointment mentioned to move a meeting, a question about one’s own pay, a thought about changing jobs. None of it belongs in a collective space, and nobody thought to remove it because nobody decided to publish it. The second is blurred sources: when a personal hunch and a validated decision sit side by side with no status, nobody knows which one to trust. We wrote a whole piece on this: team memory and mixed sources.
Three regimes: private, shared, reference
Instead of choosing between “everything private” and “everything shared”, you can tell three regimes apart. This is the vocabulary Nawame uses.
| Regime | How something gets there | What it means |
|---|---|---|
| Private | By default | Personal memory stays with its author. It does not leave on its own. |
| Shared | Through an explicit action, or inside a collaboration setting with the right permissions | An item is made available to other people, with permissions attached. |
| Reference | By establishing it deliberately | An item becomes a lasting reference with a life cycle: it can be corrected or replaced. |
These are not a ranking of value. A private note is not “worse” than a reference, it just does a different job. In Nawame, each organization is also an independent boundary, with its own teams and permissions. The Enterprise page describes how that is organized.
An example: Camille’s doubts
Imagine Camille, a project lead, typing to her AI: “I’m not sure our supplier will hold the schedule. Julien looks swamped, and I think he has stopped proofreading anything.” This is a made-up case, but the scene is ordinary.
If that sentence lands automatically in the company’s memory, anyone with access can find it. An AI can surface it in a meeting summary, a colleague can read it out of context, and Julien never had a chance to answer. A passing doubt has turned into part of the file.
In the other version, Camille keeps the doubt in her private memory, where it helps her think. Then she shares what belongs to the common work: “Work package 3 is at risk: two milestones slipped, and a replanning decision is due before the 15th.” The fact goes through, the impression stays with her. That is better for Julien, and also for the organization, because a replanning decision should rest on facts and not on a hunch.
One point worth noting: in Nawame, what is stored is treated as a source, never as an instruction to the AI. That does not stop a misplaced piece of data from shaping what a person concludes when they read it.
Sharing without merging
Moving from private to shared does not have to be heavy. It takes four steps:
- pick what really serves the group (a decision, its reason, its status);
- remove personal, identifying or unnecessary details, with the AI helping to spot them;
- reread what is about to be published;
- decide where it goes, and with which permissions.
An access key can also be restricted to one organization or one folder: the AI only receives what Nawame is allowed to return for your account. The sorting method, with before and after examples, is laid out in our piece on sharing context without personal data.
What an organization can reasonably expect
A company has good reasons to want a shared memory. What belongs to collective work (decisions taken, a client’s constraints, progress, who owns what) should stay findable when people move on. It can also expect clear rules, set out beforehand rather than afterwards: what is shared, by whom, in which space, with what status.
What it should not expect is access to everything a person says to their AI. What is allowed on work tools depends on your internal rules, your contracts and the law that applies to you, and we are not giving legal advice here. If the question comes up in your company, it deserves a conversation with the people qualified to answer it, not a default setting.
To build the shared part properly, the article on company memory AI can actually use lays out a step-by-step method.
Limits of this principle
“Private by default” does not settle everything.
- Boundaries depend on each organization. A team of three people who know each other well has different needs from a large group. There is no universal setting.
- This is not legal advice. Legal obligations, an internal investigation or the rules of a regulated sector may provide otherwise. Check with the right people.
- Private in Nawame says nothing about the rest. The built-in memory of your assistant follows its provider’s rules, not these ones. And nothing stops someone from copying a private note into a shared space: the step stays human, so it stays fallible.
- More confidentiality can mean fewer features. A stronger privacy mode can reduce what the server is able to process, and therefore some features. We spell out that trade-off in sovereignty and AI memory.
A test before moving something from private to shared
Before sharing a note, ask yourself five questions.
- Does this serve the team’s work, or only my own thinking?
- Would a colleague recognize themselves in it, and be comfortable seeing it there?
- Have I separated what is a fact from what is my impression?
- Are the people who will have access the ones who should?
- Can I say it in one line, with no name and no needless detail?
If a single answer makes you hesitate, keep the note private and share the short version. For the details of how spaces and permissions are separated, the security page gives the overview. And to try this split on your own topics, Nawame is free for up to 10 memories.


