Privacy & Portability

Your data belongs to you: what should a real AI memory export contain?

Short answer

A real AI memory export contains the memories themselves, but also the useful history, what replaced what, and where each piece of information came from. Without that, you get a pile of text, not a context anyone can understand elsewhere. Formats should also be versioned, so older exports can still be re-imported.

An opaque graphite block stays in place while a bundle of copper threads with small beads flows out of a glass case across the frame.

Almost every service promises that you can “get your data back”. The promise is reassuring, and it is often kept to the letter: you receive a file. The real question comes next. What can you do with that file in six months, in another tool, in front of a colleague who was not there?

Why an Export button is not enough

A memory export is a copy, in a format readable outside the service, of what that service kept for you. Put like that, it sounds simple. In practice, two exports can share the same name and have nothing comparable in value.

The first is a flat list of notes: sentences, in the order they were written. You can reread them, which is already something. But you no longer know which one was still true at the end, why an option was dropped, or where a claim came from.

The second keeps the thread: what was decided, what replaced what, where each piece of information came from. It weighs barely more, and it reads like a story rather than a garage sale.

The difference is not file size. It comes down to what someone judged useful enough to send out with you.

What a useful export should contain

Here is what we expect from an export, and what Nawame lets you take with you: your memory, your useful history, the replacements and the provenance.

The memories themselves

This is the minimum: decisions, reasons, constraints, observed results, open questions. If the export stops there, you have a snapshot of the final state. That is enough for some uses, such as finding the content of a note.

The useful history

“Useful” is a chosen word. Keeping every draft and every hesitation would drown what matters. What counts is what lets you understand a change of course: the earlier position, the later one, and the moment you moved from one to the other. We go into this in our article on what happens to a decision when it changes.

The replacements

A replaced piece of information is not an erased one. If you wrote “we stay with this supplier” and three months later “we are switching”, a faithful export must say that the second decision took the place of the first. Without that link, reading the export may bring the old version back up as if it were still in force.

The provenance

Provenance answers a simple question: where does this information come from? It may come from you, from a conversation with a tool, from an import, from a document. Finding a piece of information does not make it true; knowing where it came from helps you decide how much to trust it. An export without provenance forces you to start over.

Imagine a change of tools after a year

Imagine a freelance consultant who has kept a client’s context for a year: why a provider was dropped, which budget constraint weighed in, which question was still open in June. One day she wants to switch tools, or simply use a second one for a specific need.

With a flat list of notes, she spends an evening rebuilding the order of events. With an export that keeps replacements and provenance, she sees at once that the March decision was replaced in September, and that the budget constraint came from a conversation with the client, not from her own estimate.

It is not magic. It is the same content, better packed. But it is the difference between moving house and taking inventory after a fire. For picking a project back up in another tool, the method is described in how to move project context from one AI to another.

Your data on one side, the engine on the other

Two things get mixed up: your data, meaning what you entrusted to the service and what it tells; and the engine, meaning the way a service chooses, ranks and brings that data back at the right moment.

The data should leave with you. The engine stays Nawame technology: the internal mechanisms for selection and retrieval are not part of the export. We would rather say so plainly than let you believe that exporting also takes the know-how with it.

The split is defensible for a practical reason. What belongs to you is your story. How a tool finds it again is an engineering job that another tool will do in its own way. What matters is that the story arrives complete and readable.

Versioned formats, so older exports can be re-imported

An export that only the current version can read is worth little. Nawame’s export formats are versioned: you can re-import older exports, even after the format has evolved.

This spares you a classic trap: a file carefully saved a year ago that is now useless because nothing can read it.

One rule comes with re-import: imported information stays identified as such. It never becomes authoritative automatically. If you import an old export, the content comes back with its provenance, and it is up to you to decide what regains the status of reference information. An import must not be able to launder a doubtful piece of information by giving it an official look. How to sort what you import is detailed in importing memory without importing hallucinations.

What if Nawame disappeared?

This is the question to ask any service that holds your context, and we ask it of ourselves too. Our answer is one idea: the export should let you keep understanding your context without the engine.

Concretely, you keep your memories, their useful history, their replacements and their provenance in a versioned format. You can reread them, file them elsewhere, hand them to another tool. You do not reproduce Nawame’s proprietary mechanisms, and that is not the goal.

We are not saying everything would go without friction. Another tool will not have the same retrieval reflexes. But your story would not be held hostage: your tools change, your story stays.

Current limits

Several things are not solved, and it is better to read them here than to discover them later.

  • An export is only worth what was kept. Nawame deliberately does not memorize everything. If a piece of information was never retained, it does not come out.
  • The engine does not travel. Another tool will read your data, but will not bring back memories the way Nawame does. Portability of meaning is real; portability of behavior is not. The difference between getting your data out and taking its meaning with you is developed in data portability vs portability of meaning.
  • An import stays cautious. Since imported information is never automatically a reference, it may take some human work to give the items that matter their status back.
  • Formats evolve. Versioning lets you re-import older exports, but it does not guarantee that a third-party tool can read every version.
  • Nothing replaces a reviewer. A complete export does not tell you what is still relevant today. The judgment remains yours.

What this changes for you

Before entrusting your context to any service, ask it three things: what comes out in the export, in what state, and will I still be able to read it in a year? If the answer is “a text file”, you know what to expect.

At Nawame, export is available from the free plan, as is the connection to compatible AI tools; details are on the pricing page. To understand how privacy is set for what you export and what you keep, read our article on sovereignty and its trade-offs, or the security page. The general workings of Nawame are described in the documentation.

If you want to see this on your own context, you can start free.

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