Portable AI memory: can you take it with you?
You can move your photos, your contacts and your bank data. Moving what your AI knows about you is still mostly copy and paste.
Configure · · 4 min read
In short
- Portable AI memory means the things an assistant has learned about you can be brought into another assistant or app, by you, when you choose.
- Today most portability is manual: an export file or a pasted summary that is stale by the next conversation.
- Real portability goes both ways, stays current, and is controlled by the user for each product.
- A tool that only pulls memory out of an assistant is not portability. It has to serve and sync, with consent.
Portable AI memory means that what one assistant has learned about you can be brought into another assistant or app, by you, when you choose. Today the honest answer to "can you take it with you?" is: partly. You can usually export or paste a summary. What you cannot easily get is memory that follows you and stays current.
We build Configure, infrastructure to connect agents, apps and chats, and portability is the reason most people first use it. This is how the options compare.
Why does portable memory matter now?
Personalization has become a default expectation for consumer products. Control over your own data is becoming one too. People have typically spent real time talking to an assistant about something, and they want that to continue elsewhere: a plan started in one place, finished in another.
The alternative is lock-in by accident. Nobody chose their assistant because it holds their shoe size. But after a year of conversations, switching means starting over, so people stay where their history is. We think platforms should compete on the intelligence of their models and on the experience they build, and not on holding facts about you that you cannot move.
What can you actually move today?
| What you do | What you get | The catch | |
|---|---|---|---|
| Data export | Request an archive of your chats | A large file of raw history | Most products cannot read another product's archive |
| Pasted summary | Ask one assistant what it knows about you and paste the answer into another | A snapshot of your preferences | One time only. Stale by the next conversation |
| Built-in import | Use an assistant's own "bring your memory" feature, where one exists | That assistant learns your basics | One direction, into that one product |
| A connected profile | Connect your assistants to a profile you own | Memory that any approved product can read, and that keeps updating | Each product has to support it |
The first three are worth knowing because they work right now with no setup. They share one weakness: they are copies. The moment you tell either assistant something new, the copy is out of date.
What should "portable" really mean?
We use four tests.
- In and out. You can bring your memory into a product and take it back out. Portability that only flows toward one company is a funnel.
- It stays current. When one assistant learns something new, the others you have connected can know it too, without you exporting again.
- Per product, and revocable. You approve each product separately, you can narrow what it reads, and you can cut it off. Its access ends when you say so.
- It takes a button or two. Technical users can already wire this up by hand. Most people will not, so a real answer has to be simple enough for everyone else.
Why is there no "download my memory" button everywhere?
It is tempting to think the fix is for every assistant to expose its raw memory to any tool that asks. There are good reasons that has not happened. A tool whose only purpose is to pull a person's memory out of an assistant gives the user nothing inside the conversation they are having, and it turns the assistant into an unconstrained exit for personal data. That is how trust gets broken.
The version that works is different in kind. The layer in the middle does not act as a one-way pipe. It serves and syncs memory in both directions, and it does so only with explicit permission from the user, for each agent. The assistant keeps a clear boundary, the user gets the continuity they want, and every connection is one they approved. Put simply, it is the difference between explicit user consent and passive exfiltration.
Who should carry the memory between assistants?
Someone neutral. Moving information between counterparties has historically been the job of a party that is not one of them. A lab has its own model to favor and a platform has its own ecosystem to favor, so neither is a natural home for everyone's memories from everywhere else. An independent layer with no model and no consumer platform of its own can serve all of them evenly. We go further into this in Plaid for AI: why agents need a connect button.
How does it work with Configure?
From the user's side it is a short flow. A product you are trying shows a connect step. You pick the assistants and apps you already use, approve, and return to the product, which can now start from what you have already shared. If an assistant cannot connect directly, you can bring existing memories in as a one-time import. An import adds what you had. A connection is what keeps it current.
Afterwards you manage everything from your profile: what each product can read, which apps are connected, and what each agent has added. You can remove any of it, or delete the profile.
From the developer's side it is one integration. Your agent reads the user's approved memories before the first message and can save what it learns back to the same profile. The quickstart walks through it end to end.
What should you do today if you just want to switch assistants?
Ask your current assistant to write out everything it knows about you that would help a new assistant: your work, your preferences, ongoing projects, how you like answers written. Read it, delete anything you would not want carried over, and paste it into the new one. It is a snapshot and it will drift. For a one-time move it is perfectly good, and it costs five minutes.
When you find yourself doing that for the third product, that is the point where a connected profile starts to pay for itself.
Add a connect button to your product
The quickstart takes one page, end to end: a chip in your app, Configure's hosted page, and your agent reading the user's context on the first turn.
Read the quickstartNext: AI onboarding without the questionnaire