How to move from ChatGPT to a multi-model app without starting over

Nobody moves to a multi-model app the way they move a phone. There is no backup file to restore and no app that can log into your old account and pull everything through. What moves is the part that made the old app feel like yours: the standing instructions, the corrections you made a hundred times, the details it learned about your work. That part travels in about twenty minutes. The transcripts do not travel here at all, so keep the old archive if you may need to refer to it. (Gemini now imports chat-history exports, though not in the EEA, the UK or Switzerland. MultiChats does not.)

What you give up, said first

If your working day runs through a custom GPT you built, that does not come with you, and OpenAI plans to retire custom GPTs on 11 December 2026 in favour of its plugins. The nearest thing here is a folder on Pro or Pro+: memories attached to it carry your instructions into each new chat inside, and its files go with those chats. That covers a GPT made of instructions and reference files, and nothing that calls another service. Anything wired into your Google or Microsoft account stays behind too. We do not connect to your inbox, your calendar, your drive or your spreadsheets, we do not join or transcribe meetings, we do not run code, and we do not generate video.

If one of those is load bearing for you, stay where you are. Somebody whose custom GPT connects to other services has a worse week after moving to a general chat app, and no amount of model choice fixes it. The rest of this guide is for everyone else: people who mostly type questions, paste documents, and want more than one model answering them.

Step 1. Sort what travels from what does not

Two lists, five minutes. Things that travel: your tone and formatting preferences, the standing instructions you repeat in every chat, your job and the projects you keep referring to, the corrections you have made so often the assistant learned them.

Things that do not: conversation history, uploaded files, images you generated, shared chat links, custom GPTs, and anything specific to that lab's own products. Assume all of it stays behind and keep the old account open for a month as a read only archive. That is easier than trying to salvage it.

Step 2. Ask your old assistant to write down what it knows about you

This is the trick that makes the whole thing work: the assistant that stored your memories can be asked to print them.

On the web, open Settings, go to the Memories tab and choose Import from AI. In the iOS or Android app, go to Profile, then Memories, then Import. Pick your old assistant from the list. Six are covered: ChatGPT, Claude, Gemini, Grok, Perplexity, and Mistral's, listed under its old name, Le Chat.

You get a ready made export prompt with a copy button. It asks for every stored memory, one entry per line, under six headers, with your original wording preserved rather than paraphrased. Paste it into a fresh chat with the old assistant and copy the block it gives back.

Do not accept a summary. The prompt ends by asking the assistant to confirm the list is complete or say how many entries remain. If it answers with a tidy paragraph instead of a list, ask again for every entry individually. A summarised export loses exactly the specific details that made the assistant useful.

Step 3. Paste it in and check the split

Paste that block into the next screen and press Parse memories. The text goes to a model that breaks it into discrete entries, one per fact, with a category label taken from the headers.

You then get a preview list with everything ticked. Untick anything you have outgrown, which is usually more than you expect: assistants hold on to projects you finished last year. Press Import and the selected entries are saved, each tagged Imported plus its category.

Three limits to know before you paste:

The pasted text caps at 50,000 characters. A long export should go in two passes.
Up to 100 memories import in one batch.
Any single entry longer than 10,000 characters, or with a title over 200 characters, is shown in red and skipped automatically so it cannot break the rest of the batch.

Saved memories are a paid feature here. The parse step returns a subscription notice on a free account. Pro is $20.99 a month or $199.99 a year and holds up to 200 memories, Pro+ is $34.99 a month or $349.99 a year with no cap, and both are in the pricing section. If you are not paying yet, Step 4 is your version of this whole guide.

The file import button is narrower than it looks. The other option on that screen on the web, Import files, takes .md files up to 1MB each and nothing else. It will not read a ChatGPT export archive, a JSON dump or a PDF.

Step 4. If you are not paying, one box does most of this

There is a free landing place for everything you just exported, and it carries most of the value.

On the web, open Settings, then the Account tab, and find Personal preferences. On the phone it is Profile, then Personalization, under the heading "Anything else the AI should know". It takes 5,000 characters and attaches to every conversation you start, on every model, with no plan gate.

Paste your export in, edit it down to what still matters, and you have the personalization without the memory system. What you miss is granularity: saved memories can be switched on per chat and per folder, while this box applies everywhere or nowhere.

Step 5. The star is what turns a memory on

Imported memories arrive switched off, and this is where new arrivals get confused.

In the Memories tab, each entry has a star. Starred means "include in all new chats by default", which is what people assume the import already did. It did not, deliberately: a hundred imported memories firing at once is worse than none.

Star the ten or so that describe how you want to be answered. Leave the project specific ones unstarred and attach them to the chat or the folder that needs them.

Step 6. Folders on day one

Folders are free, up to 25 of them, and they nest. Make one per live project on day one and attach memories to the folder rather than to individual chats, so anything you start inside it inherits the right context. Resist recreating your old history: start what you are working on this week and let the archive be an archive.

Step 7. Three things that will feel wrong for a week

One button starts a new thread, the other overwrites your answer, and web and phone swap the names. On the web, Try Again on any answer branches: it creates a new thread carrying the conversation up to that point, with the model you pick, leaving your original answer intact. On iOS and Android, Try again is the destructive one, replacing the answer in place, and Branch here is the button that creates the new thread. Learn which you are pressing before you press it on an answer you like.

Answers arrive one at a time. Branching gives you a second thread in the sidebar. There is no split screen, and nothing appears next to anything else.

Model choice becomes a decision you make. 60 models from 18 providers, 21 of them on the free plan, and no single house style to get used to. Changing model part way through a conversation is a topic for another day.

One last check, since you are about to paste a personal export somewhere new. We do not train on your conversations. Some of the providers behind those models might, depending on which one you pick, and where we know of a difference the model's info card carries a notice. Six of the 60 do, and Grok's is unambiguous: "xAI may use your conversations to train Grok models."

So the last step of the move is the smallest. Check the model, then choose where your export lands.