Privacy & Portability

Sovereignty and AI memory: why privacy can require trade-offs

Short answer

The more a piece of content is protected from server-side processing, the less the service can do with it. Nawame’s sovereign mode strengthens confidentiality: some content stays unreadable to the server, which reduces certain capabilities. It is available from the free plan, and you decide where to set the dial.

Two glass cubes, one clear and one frosted, each threaded by a copper line, joined by a fine balance line.

“Your data is private and the service stays fully smart”: that sentence is pleasant to read, and it hides a trade-off. To help you find something, a service generally has to understand what it is holding. The more you forbid it from understanding, the less it can help. This is not a weakness specific to one provider. It is a real tension, and it is better to face it.

Why this is not a setting like the others

Sovereignty, in the sense we use it here, means the control you keep over your content: where it is hosted, who can read it, what can be done with it. Those three questions are not settled by a single switch.

Take the simplest example. To connect today’s situation to a decision made long ago, a service has to read both and judge that they resemble each other. That is what a semantic search does: it compares content by meaning rather than by exact words. If the service cannot read the content, that comparison is no longer possible as it stands.

This is why no serious service can promise total confidentiality and every feature at once. We do not promise it either.

Personal memory, private by default

Let us start with what does not change from one mode to the other. Your personal memory is private and encrypted on Nawame’s infrastructure, hosted in Europe, under your control. Access depends on your account and on the rights you grant.

It never automatically becomes a team or organization memory. When a personal context needs to become shared, you choose to share it, and the AI can help remove personal, identifying or unnecessary elements before publication. Going from private to shared is a deliberate act, never a side effect. How to prepare context for sharing is detailed in how to share AI context without sharing personal data.

For many uses, that is enough. We come back to this below.

What sovereign mode changes

Sovereign mode is a mode of reinforced confidentiality: some content stays unreadable to server-side processing. In other words, even the automated processing the service performs cannot read that content.

The benefit is easy to state: less of your content is exposed to the service’s processing. If your context touches sensitive subjects (a delicate personal situation, a confidential file, a strategy that must not circulate), this mode answers a legitimate worry.

One point matters: sovereign mode is available from the free plan. It is not a premium option. Reinforced confidentiality is not a privilege reserved for those who pay. You can check the pricing page for the details of each plan.

What sovereign mode costs

Content the server cannot read is content the server cannot work with. That reduces certain capabilities, the ones that require understanding the content on the server side. Semantic search is the most telling example: what a service cannot read, it struggles to connect to anything else.

We do not claim this cost is negligible. Depending on what you protect and how you work, you will feel it more or less. A highly structured context that you know how to find yourself will suffer little. A vast context, where you count on the service to bring up the right memory at the right time, will suffer more.

That is why the wording “maximum confidentiality, full features, no compromise” is not ours. It would be false, and a well-informed reader would spot it at once.

When the usual way is enough, and when sovereign mode is justified

Nobody needs the same level of protection for everything. A few markers, to be taken as leads rather than rules.

  • The usual way fits when your context is everyday work, you mostly want it retrieved well, and the protection offered by encryption, European hosting and access rights is enough for you.
  • Sovereign mode is justified when the content is sensitive enough that confidentiality comes before search comfort, or when you want to reduce what server-side processing can read.

A well-kept document, a notebook or a tool’s native memory can also be enough if your need is modest (the trade-offs between these families are compared in local memory or independent memory). We do not want to sell you a level of protection you have no use for.

Imagine a consultant and two files

Imagine a consultant who works with two clients. The first file is about organizing a monthly workshop: routine decisions, not very sensitive, which he wants to see resurface when a topic comes back. The second is a confidential negotiation that he does not want exposed to any automated processing.

For the first, the usual way seems right to him: he wants the full benefit of search help. For the second, he leans toward sovereign mode, knowing that the service will help him less in connecting that content to other items. If he goes that way, he does it knowing the price. The trade-off is his, and it does not have to be the same for every subject.

Organizations: independent boundaries

The same reasoning applies to teams. Each organization is an independent boundary: rights, teams and memories are managed there separately from other organizations, and separately from your personal memory. What belongs to one does not mix with the other.

Inside an organization, there is also authoritative context: information that has the status of reference, such as a business rule, a policy or an official decision. Authority concerns the status of the information, not the idea that it could never be deleted. For the business framework, the enterprise page describes the approach. The boundary between personal notes and shared memory is discussed in why personal memory must not automatically become company memory.

Current limits

  • No absolute promise. Sovereign mode strengthens confidentiality; it does not turn Nawame into a vault with no trade-off. Some capabilities are reduced, and we say so.
  • The trade-off depends on you. We cannot decide for you where to set the dial. Misjudging your own need, in either direction, remains possible.
  • Retrieving less can be costly. If you count on the service to bring back an old trade-off, content that is unreadable to the server will be harder to connect to a new context.
  • Confidentiality is not the only question. Well-protected content can also be badly written, outdated or wrong. Being retrieved or protected does not mean being true.
  • This article is not an audit. For the security framework, see the security page. We do not detail internal workings here.

What this changes for you

Before choosing, ask yourself one question: what would you lose if this content were read by automated processing, and what would you lose if it could no longer be? The answer depends on each file, and it can change over time.

Know also that you keep control over the way out: export lets you take your memory with you, as our article on what a real export should contain explains. And if you wonder what a tool’s native memory can offer alongside, we covered it in a dedicated article.

If you want to try it, you can start free, sovereign mode included.

Your next prompt should not have to contain your whole story.

Start building a memory your AI tools can find again when it matters.

Start free