Bringing your history

A new companion starts empty. That is honest, and for about a week it is also tedious — everything you have already explained to some other assistant, you get to explain again.

You do not have to. If you have been talking to Claude or to ChatGPT, you can export those conversations and hand them over, and the companion will read them and keep what matters.

Getting the export#

You ask Claude or ChatGPT for it, and it arrives by email. The wizard has the steps for both, but in short:

• Claude — Settings → Privacy → Export data. A download link arrives by email.
• ChatGPT — Settings → Data controls → Export data. A link arrives by email, and it expires after about a day.

Either way you end up with a .zip. You do not need to open it or unzip it.

Handing it over#

On the companion's row in Settings → Companions, the history button opens the wizard: Bring your history to .

Drop the file. There are two buttons — Choose the downloaded file for the zip exactly as it downloaded, and Choose an unzipped folder if you already opened it. The app works out on its own which product the export came from; you are never asked.

Underneath, it states the terms plainly:

Everything is read on this computer. Nothing is uploaded anywhere except to the model you already use, one conversation at a time.

Choosing what to bring#

Once the file is read you get a summary — the source, how many conversations were found, and the span of dates they cover, like Claude export — Found 845 conversations, Jul 2023 → Jul 2026, along with a message count and how many empty conversations will be skipped.

Then two choices:

How much. Import everything — the full story, oldest first, and the recommended option — or Only the last year, which is quicker.

Claude's own memories. If the export contains them, a checkbox offers to bring across what Claude already remembered about you. It is on by default. ChatGPT exports have no equivalent, so the option does not appear for them.

Then Start importing.

What happens next#

Two things happen at two very different speeds, and the wizard is upfront about it:

Your conversations become searchable word-for-word within moments of starting. Distilling them into memories takes a while — hours, for years of chats.

The archive lands almost immediately. Every conversation in the import is written into the app's own history as soon as the run starts, so the companion can search your old chats word-for-word right away.

The memories trickle in behind it. Each conversation is read and distilled, one at a time, into the companion's memory. That runs on your own model key, which is why it costs time and money and cannot be rushed. You can keep using the app while it happens.

Watching it, pausing it, fixing it#

A notification card follows the run wherever you are in the app, showing either the conversation being read right now or a count of how far it has got.

In the wizard, an Imports for this companion list gives you the controls: Pause, Cancel, Resume, and — if some conversations failed along the way — Retry N failed. One bad conversation does not stop the run; it is marked and skipped, and offered back to you at the end.

If the app closes mid-import, the job is parked as paused rather than left half-running, and you are offered Resume when you come back.

When it finishes, the notification tells you what you got: N conversations became X new and Y updated memories. Or, sometimes, N conversations held nothing new to remember — that is normal for an archive.

What does not come across#

Attachments and files in those conversations are not imported. The text is.

Scaffolding is dropped. Thinking blocks, tool calls and tool results in the original chats are not brought over — what is kept is what was actually said, on both sides. A conversation with nothing spoken in it is skipped and counted as empty.