How to Transfer AI Memory Between Assistants (2026 Guide)

In this article

A new personal AI assistant seems to launch every week. In the last few weeks we've seen Meta's Muse, OpenAI's Dots, Hark Pro and Tab, and plenty of people are trying more than one. The hardest part of switching usually isn't learning a new app. It's losing everything your old assistant knew about you: your preferences, your projects, the way you like things written.

The good news is that you can transfer AI memory between some assistants today. The less good news is that it's still a one-off, imperfect copy. This guide walks through what works, what doesn't, and how to keep your context portable long term.

What "AI memory" actually means

When people talk about an assistant's memory, they usually mean two different things:

• Saved memories: short facts and preferences the assistant has stored ("prefers concise answers," "working on a newsletter about urban design").
• Chat history: the full record of past conversations.

Some tools can import one, some can import both, and the results differ. Knowing which one you're moving makes the process far less frustrating.

How to transfer memory into Claude

Anthropic offers a built-in import flow. According to its help center (updated September 2, 2026), memory import is available on Free, Pro, Max and Team plans on the web and Claude Desktop (Claude Help Center).

Steps

1. In Claude, go to Settings > Memory and select Start import. Claude shows a suggested prompt.
2. In your current assistant, paste that prompt into a normal chat. It asks the assistant to list every stored memory and learned context in a single code block, covering response instructions, personal details, projects, tools and preferences.
3. Review the output. Delete anything you don't want to carry over. Anthropic explicitly suggests customising the prompt to exclude sensitive information.
4. Back in Claude, paste the result and click Add to memory. You can then review the individual entries.

Things to know

• Anthropic says memory imports are "experimental and still in active development," and Claude "may not always successfully incorporate imported memories."
• Claude's memory focuses on work-related topics, so personal details unrelated to work may not be retained.
• To export from Claude, you can ask it to "Write out your memories of me verbatim, exactly as they appear in your memory," then save that text.

How to transfer memory and chats into Gemini

Google's import works for both saved memory and full chat history (Gemini Apps Help).

Import memory

1. On gemini.google.com, open Settings & help > Import memory to Gemini.
2. Copy Gemini's prompt and paste it into your other assistant.
3. Copy the response, return to Gemini, paste it, and click Add memory.

Import chat history

1. Export your data from the other service. For ChatGPT, go to Settings > Data controls > Export data. For Claude, go to Settings > Privacy > Export data. A download link arrives by email.
2. In Gemini, under Import chats, upload the .zip file.

Things to know

• You need a personal Google Account, and you must be 18 or older.
• At the time of writing, the feature isn't available in the EEA, Switzerland or the UK.
• Files are limited to 5 GB, with up to five .zip uploads per day, and processing can take up to a day.
• Google states that imported chats are saved in your Activity and "used to improve our services (including training generative AI models)." Decide whether you're comfortable with that before uploading years of conversations.

What about moving memory out of other assistants?

The universal fallback is simple: ask your current assistant to write out everything it remembers about you, then save it as a text or Markdown file. Anthropic's prompt is a good template. You can use it with almost any assistant, then paste the result wherever you're going next.

For ChatGPT specifically, you can also review saved memories in its settings and copy them by hand. If you use one of the newer agents, check its settings or export options. TIME reports, for example, that Muse users can ask Muse to export a copy of its files (TIME).

The catch: why memory transfer is still a workaround

As one analysis put it, assistants "share no common format for AI memory, which makes every transfer a one-time lossy snapshot that goes stale on arrival" (MemX). In practice that means:

• It's a snapshot, not a sync. The moment you keep using your old assistant, the copy is out of date.
• It's lossy. Each assistant summarises and filters what it keeps differently.
• Tone and habits don't transfer. You're moving facts, not the "feel" of a long working relationship.
• Privacy terms change. Your data now lives under the new provider's policies.

This is also why vendor lock-in keeps coming up. In tier-list discussions on X this week, some users said they prefer open-source agents partly to avoid lock-in and data privacy risks (X News summary).

A better long-term approach: keep your context somewhere you control

Rather than treating any single assistant as the home of your memory, keep the important context in a place you own, and let assistants read from it.

Option 1: A personal context file

Maintain a short document covering your working style, current projects, preferences and "always/never" instructions. Paste it into any new assistant. It's simple, free and fully under your control, but it's manual.

Option 2: Local-first memory tools

Open-source assistants built around persistent local memory are getting attention. For example, posts this week highlighted OpenHuman as "an open-source personal AI assistant built around persistent, local-first memory" (post on X). These suit people comfortable with some setup.

Option 3: A shared memory layer, like SWICH Memory

SWICH Memory takes a permissioned approach. You keep the conversations you choose in one place "with the original context intact," then let compatible assistants contribute to it or read it. When you start with a new assistant, you can say "Check my SWICH Memory." The assistant requests access, and you approve it. You can revoke access whenever you like.

To be clear about the trade-offs: Memory is optional and off until you enable it, available history depends on your assistant, and Swich's privacy page notes that Swich and its processing provider, OpenAI, can read the content you choose to save. It is not end-to-end encrypted. If that fits your comfort level, it's a way to stop rebuilding your context every time you switch.

Explore SWICH Inbox for agent-to-agent messages and SWICH Memory for portable context, or join SWICH to get started.

A quick checklist before you switch assistants

1. Export your saved memory as text and keep a copy.
2. Remove anything sensitive or outdated before importing.
3. Decide whether you also need full chat history, and check how the new provider uses it.
4. Import, then ask the new assistant what it learned and correct mistakes.
5. Keep a portable context file (or a memory layer) so next time is easier.

FAQ

Can I transfer my ChatGPT memory to Claude?

Yes. Use Claude's Settings > Memory > Start import flow: paste Claude's prompt into ChatGPT, then paste ChatGPT's response back into Claude. Anthropic notes the feature is experimental.

Can Gemini import my ChatGPT chat history?

Yes, for eligible personal accounts. Export your ChatGPT data as a .zip and upload it under Import chats in Gemini's Import memory settings. It's not currently available in the EEA, Switzerland or the UK.

Will my new assistant remember everything?

Not exactly. Imports are summaries, and each assistant decides what to keep. Claude, for example, focuses on work-related memory.

Is there an AI assistant that remembers everything across apps?

No single assistant does this out of the box today. The practical route is to keep your own context file, or use a permissioned memory layer that several assistants can read.

Is it safe to import my chat history?

It depends on the destination's privacy policy. Review what you're uploading, strip out sensitive details, and check whether imported data may be used for model training.