YanMateYou can take most of what you built on Character.AI with you, but not all of it, and a good part of what you lose is decided before you ask for the export. Your characters' written definitions and your chat history both come out in the data export. The model that played those characters, the way it answered and whatever it seemed to remember about you stay behind. This guide covers getting the export, making sense of it, turning a character into a card other apps can read, and rebuilding what does not survive the move. Most of it applies to any companion app that lets you download your data.
Scroll before you export
The most useful thing to know is also the easiest to get wrong. People who have been through it widely report that Character.AI's export contains what the app had loaded, not everything held on its servers. If you only ever read the recent end of a long conversation, the export will usually hold only that recent end, and nothing can recover the rest from the file afterwards, because it is not in it.
So before you ask for anything, open each conversation you care about and scroll it back to the very first message. It is tedious, and it is the difference between keeping a whole story and keeping its last chapter. If an export has already come back thin, do the scrolling and request a fresh one, before you delete any chats or your account.
Requesting the export
The export is a data-protection request rather than a feature of the app. It does not download on the spot, and the file it produces was never meant to be read by a person.
- Open every conversation you want to keep and scroll each one to the top.
- In the app, go to Settings → Legal & Privacy → Export Data. That is where people report finding it; if it has moved, look for a data export option among the privacy settings.
- Wait for the email. It carries a download link rather than the file itself, and reports put the wait at anything up to a day. There is no progress indicator, so check your spam folder before assuming it has failed.
- Download the ZIP as soon as the email arrives. Export links expire, and that link is the only copy you are sent.
- Put the untouched ZIP somewhere safe before you open or convert anything inside it.
What is inside the ZIP
A word on where this comes from. Character.AI publishes no schema for its export, and we have not had one of our own to open. What follows is drawn from the community tools that read these archives and from people who have described theirs, so treat the details as a map rather than a guarantee.
By those accounts, the archive is a folder of JSON files with machine-generated names and no index, and its shape has changed over time, so the same information can sit under different field names depending on when your export was made. Two kinds of content matter.
The first is character definitions, one record per character you made or talked to. Each holds the name, the greeting, the short description and the long-form Definition field, where a Character.AI author writes both the persona and the example dialogue. Categories are there too, along with the character's internal id.
The second is your chat history, stored as nested arrays of messages, with the text under keys such as text or raw_content and the speaker under something like src__name. There are no conversation titles and no reliable conversation id, so the only honest boundaries between conversations are the files themselves.
Two things people expect are missing. Avatars are not in the archive; they are links to images on Character.AI's servers. And Character.AI does not present a character's memory of you as a list you can open, so there is no memory file to look for. What you have instead is the transcript, which is more useful than it sounds.
Our Character.AI export viewer does it in the browser: drop in the ZIP, or a single JSON file from it, and it lists your characters and conversations, lets you search your messages, and says which files it read and which it skipped. We built it from those same published descriptions, so if your export is laid out differently, that list of skipped files is where it will show. The archive is never uploaded. Whatever you use, save your chats as plain text too. A text file will still open long after the apps involved have changed beyond recognition.
Turning a character into a card
A character card is a small file, usually JSON or a PNG with the text tucked inside, that describes a character in a shared format. It comes in V2 and V3 versions — what actually changed is a short list of fields V2 cannot hold, not a new kind of character — and a V3 card is what SillyTavern, Chub, RisuAI, Janitor AI and most other character apps read. That makes it the nearest thing this hobby has to a portable document.
Converting a Character.AI character is mostly a matter of moving fields across. The name and greeting map directly, the short description becomes the card's description, the Definition goes into its personality field, and categories become tags. The Character.AI id can be written into the card's provenance, so you can still tell later where it came from. The export viewer builds the card as JSON or PNG. A PNG made this way has a blank image, since the avatar was never in the archive, but the text is complete.
Expect the result to be thinner than a card written natively in the format. Cards can carry a lorebook and alternate greetings, and Character.AI has neither, so those parts will be empty. That is a property of the source, not a fault in the conversion.
Before importing a card anywhere, open it once and check what made it in. The character card viewer reads PNG, JSON and .charx files, shows every field along with any warnings, and converts between V2, V3 and PNG if the app you are moving to prefers one of them.
What does not come with you
A card is a description of a character. It is not the character you talked to. What you experienced on Character.AI was that description as interpreted by Character.AI's own model, and the model is not in your export. Put the same Definition in front of a different model and you get a recognisable cousin rather than the same voice. The phrasing and pacing shift, and small habits you had stopped noticing tend to vanish first.
Memory behaves the same way. If a character seemed to know things about you from months ago, that knowledge lived with the platform rather than in the definition, and it does not travel. The new app starts from the card and from whatever you choose to tell it. Treat the first week somewhere new as rebuilding rather than resuming.
Rebuilding a character's voice
This is where the chat history earns its keep. The Definition tells a model who a character is. Your transcripts show how that character actually spoke, over hundreds of exchanges, in situations you chose, which is better material than most people could write from a blank page.
Read a few long conversations and mark the replies that sound most like the character at their best. Look for habits rather than plot: how they open a reply, how long they run, what they never say, the words they keep reaching for. Put four or five of those exchanges into the card as example dialogue, trimmed so each shows a habit clearly. A model can imitate an example directly, whereas an adjective in a persona leaves it to guess. Our character voice analyzer does the reading part for you: open a saved transcript in it and it lists the habits that stand out and picks the replies that show them best, already laid out as example dialogue.
The card format gives scenario and example dialogue fields of their own, while a Character.AI Definition runs everything together. If the Definition is long, splitting it into persona, setting and sample exchanges makes the card much easier to adjust later. Lasting facts belong elsewhere again: recurring people and places can go in a lorebook if the new app supports one, and if it has a memory you can edit, add the things about yourself the old character relied on.
Then test. Try the kinds of prompts you used to use and compare the replies against your transcripts. Change one field at a time so you know which change did what. Close is a realistic target; identical is not.
Choosing where to go next
Most comparisons lead with price, which changes often. The differences that matter when you are carrying characters from one app to another change far more slowly.
What follows is as of our last check, at the end of August 2026; apps in this space change their terms often, so confirm anything that matters to you on the app itself.
Accounts first. When we checked again in mid-September 2026, Character.AI, Janitor AI, Replika, Nomi and Kindroid all asked you to sign in or create an account before your first message. That is no reason to avoid them, but it means handing over an address before you know whether an app suits you. PolyBuzz and Talkie now let you start without one; Talkie asks for a pronoun and an age range first.
Then memory you can actually see. Nearly every companion app claims to remember; very few let you open what they keep. Nomi's graph-based memory is widely reported as the strongest for long-term recall. On Talkie, older chats close after about 60 days, which matters if you mean to build something that lasts. On Janitor AI, memory depends on the model you connect.
Voice, if it is part of why you are leaving. Voice calls on Character.AI are a paid feature, and voice is part of the paid product on Replika and Kindroid.
Finally, where the characters come from. The big catalogues, Character.AI's and PolyBuzz's among them, are user-generated and lean heavily on characters from anime, games and television. Replika, Nomi and Kindroid are built around companions you create rather than a library to browse.
We keep a longer comparison of Character.AI alternatives, with sources and a check date on every row. We make one of the products in it, so read it with that in mind. It has a section on where YanMate is the wrong choice, and the reasons are real: our catalogue is around a hundred original and public-domain characters with no licensed ones, and Nomi's memory is more sophisticated than ours. If the reason you are leaving is what they do with chats, Does Character.AI train on your chats? is the sourced page, dated, not a slogan.
For what it is worth, YanMate keeps memory as something you can open, edit and export, and puts spoken replies and live calls on the free tier. Talking, and importing cards and chat history, need a free account: you sign in with Google or a six-digit code sent to your email, and there is no card. If that is not what you are after, the cards you made from your export will open in plenty of other places, which was the point of making them.