Comparisons

Local memory or independent memory?

Short answer

No option wins everywhere. Built-in memory is the simplest, local files give you control, a home-made base fits your own habits, and an independent memory follows your work from one AI to another. Local is not safer by default: it depends on backups, devices and sharing. Choose based on how many AI tools you use and how long your projects last.

Three glass containers of different shapes on a light tray: one sealed, one resting on a stone base, one open and joined by a long copper thread to a fourth that floats above.

Picture Camille, a freelance consultant. For two months she has been planning a client’s website rebuild with an AI assistant. She has settled plenty of things: the section names, the publishing tool, what will not be migrated. Then she switches assistants and finds that none of it came along. So where does the memory of her work live, and who does it belong to?

There are at least four answers, and none of them wins every time. This article goes through each one with its benefits and its costs, Nawame included, since it sits in the fourth group. If you use one AI for short tasks, the first option is probably enough and you can stop reading here.

Four families, four trade-offs

To keep track of work done with an AI, you can:

  1. use the memory built into the tool, the one the assistant keeps by itself;
  2. keep notes locally, on your own computer, in files you control;
  3. build a home-made external base: a shared document, a wiki, a spreadsheet, a notes app;
  4. use an independent memory that travels from one AI to another.

These groups overlap in real life. Plenty of people use the first for their preferences and the second or third for their projects. The useful question is not “which is best?” but “which one covers what I am afraid of losing?”

What each one gives you, and what it costs

Memory built into the tool

Its big advantage is that it is already there. No setup, no new habit. According to the help centers of Anthropic and OpenAI, consulted on 30 September 2026, both assistants offer memory features, with settings to look at them or limit them. These features change fast, so read the official pages rather than this article for the details.

The trade-off: it belongs to one product and one account. Switching assistants means starting from an empty memory, unless the tool offers a transfer. We cover that case in why not simply use ChatGPT or Claude memory.

Local memory

Plain text or Markdown files on your computer: you see everything, you edit everything, nothing depends on a vendor. It is the most direct way to stay in charge, and a text file stays readable for a long time.

It takes discipline, though. You write, date, replace and sort everything yourself. And a cloud-hosted AI cannot read your hard drive. As an example, xAI’s documentation, consulted on 30 September 2026, says Grok rejects local or private addresses for its connectors. In practice, you paste the right excerpts into the conversation yourself.

A home-made external base

A team wiki, a shared document, a notes app: this is documentation territory, and it is very good at it. Several people can write in it, page history often exists, and the tool is more or less yours.

The trade-off is that a notes base keeps what you put in it, not necessarily the link between a decision, its reason and what replaced it. We look at this in our comparison of Notion, Obsidian and Nawame. And someone has to keep it up to date.

Independent memory

It lives outside the assistant and plugs into several. You change AI, the history stays. This is Nawame’s family, and we come back to it below. Its costs are real: an account, a connection to set up, and a reliance on its service.

A table to compare them

These are general tendencies, not measurements. Any specific product may work differently.

Criterion Built into the tool Local Home-made base Independent
Control Vendor settings Full High Rights, plus your choice of what goes in
Portability Depends on the tool Good with open formats Good for text Export provided, engine excluded
Effort Low Medium to high Medium to high Medium (account, connection)
Sharing Within the product Up to you Often easy Private by default, explicit sharing
Backup Vendor side Your job Depends on the tool Hosted in Europe, export
Dependency Strong on one vendor Low On the notes tool On Nawame and on the AI’s connection

The myth that local means safe

Local feels reassuring, and it has real strengths: your files never leave your machine. But “safer” comes down to very concrete questions.

  • Backups. A failed drive takes a year of project memory with it if nobody made a copy.
  • Devices. A lost or shared laptop exposes whatever is on it.
  • Sharing. Emailing a folder so someone else can pick up the project means leaving your own perimeter.
  • Versions. Three copies of the same file, only one of them current, is the worst case.

The reverse also holds: a hosted service is not dangerous on principle. It depends on who can reach it and under what conditions. Both deserve an honest look, not a reflex.

Where Nawame fits

Nawame is an independent working memory, hosted in Europe, that you connect to several AI tools through MCP. It keeps decisions, their reasons, discarded options, results and changes over time. Personal memory is private by default, and sharing takes an explicit action. You can export your memory and its essential structure. The details are on the security page.

Its trade-offs, plainly:

  • you need an account (created at first connection, with a code sent by email);
  • the AI has to be connected, and Nawame is called from the AI’s cloud, not from your machine;
  • sovereign mode strengthens privacy, but it can reduce some server-side functions;
  • the engine (ranking, context selection) is not necessarily part of an export;
  • it does not spare you from deciding what deserves to be kept, and sometimes the right move is to keep nothing.

On exports, our article on what a real memory export should contain describes what you can reasonably expect from any service.

Limits of this comparison

Four families is a convenient split, not a sharp one. Tools blur the lines: a notes app can plug into an AI, a built-in memory can offer an export. Do not expect your case to fit a box exactly.

We also did not compare specific products, and we measured nothing: the cells in the table are tendencies. Assistant features change quickly, and what we cited dates from 30 September 2026. Finally, we have a stake in the fourth family, so weigh our view accordingly, and see how portability plays out in practice in data portability versus portability of meaning.

Which one fits your situation

Four scenarios to help you decide:

  • One AI, short tasks, stable preferences. Built-in memory is enough. Adding a tool would only complicate things.
  • A solo project you want to keep in your own hands, with no sharing. Well-kept local files (a project status page, a decision log) work very well, as long as you back them up.
  • A team that already documents. Your wiki is still the right base. The real question is whether decisions and their reasons can be found in it.
  • Several AI tools across projects that last months. This is where an independent memory pays off: the history does not stay locked inside the first tool.

Here is a quick test. Take a real project and ask yourself three things. What would I lose if I switched AI tomorrow? Who can read this information today? Once I have forgotten, where will I go to find the reason for this choice? The answers often show which of the four families you are missing.

To try the fourth on a single topic, Nawame is free for up to 10 memories.

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