---
title: "ChatGPT, Claude or Gemini"
description: "Give your twin the facts an assistant already knows about you."
source: https://hiy.ai/docs/from-your-assistant
---

# ChatGPT, Claude or Gemini

If you've spent a year telling ChatGPT, Claude or Gemini who you are, that
assistant already holds a compact description of you. This is how you hand
that description to your twin — not by connecting anything, but by pasting it
in, once.

## What this actually is

There's no way for hiy to reach into an assistant's memory directly. No
assistant exposes it, and none offers an API for it. What arrives is a paste:
you open the assistant's own memory screen (or ask it to write out what it
remembers), copy the text, and bring it here. That's the mechanism, not a
workaround.

Think of it as a seed, not a migration. It gives your twin a head start on
facts an assistant already has about you. It doesn't carry over how you'd
phrase things, and it doesn't teach your twin to sound like you — more on that
below.

## Two ways to capture it

**Copy it from settings** when you can. Open the assistant's own memory
screen — Personalization in ChatGPT, Capabilities → Memory in Claude, Personal
context → Saved info in Gemini — and copy exactly what's shown there. What you
paste is what the assistant actually holds. This is ground truth, and it's the
option hiy offers first.

**Ask it to write the list out** when there's no settings screen to copy, or
you'd rather have it in one clean block. hiy gives you a prompt to paste into a
new chat, asking the assistant to list everything it has stored, verbatim. But
this is the assistant generating prose about itself, not reading a database —
it can include things it never actually stored, alongside things it did. This
is why the review step below exists, and why it's worth reading every line
before you approve it.

## Review before anything is stored

Pasting your text splits it into individual facts — one per bullet, per
numbered item, or one per line if there's no list. Each fact gets its own
checkbox, ticked by default. You go down the list and untick anything that's
wrong, private, or about someone else. You can leave a fact out, but you can't
rewrite one: what survives review is still what the assistant said, not an
improved version of it.

Only the facts you leave ticked are ever sent anywhere. Splitting the paste
into facts and reviewing them both happen in your browser, before any network
request — a fact you untick never reaches hiy's servers at all.

Under the list, a consent checkbox names what a memory dump can contain:
health, sex life or sexual orientation, politics, religion, finances or
relationships. An assistant's memory of you can genuinely include any of
these, so you confirm you've read what you're about to add before the button
unlocks. See the [privacy policy](/privacy) for what that consent covers and
how to withdraw it.

There's a cap of 200 facts per import, 600 characters each — plenty for a
memory dump, and a limit on how much unreviewed text one paste can turn into.

## Answers, not voice

This is the part worth being precise about. An imported memory is a summary
written by another assistant, in its own third-person register — something
like *"the user prefers concise answers"* is knowledge, not something you'd
say out loud. So it feeds your twin's answers, the same as any other source,
but it's deliberately excluded from the material that shapes how your twin
sounds. That's a real distinction, not a smaller one: if your twin is public,
anyone who chats with it can be shown one of these facts as an answer — quoted,
with this import's title next to it as the source. "Answers, not voice"
describes what the import can't do, not who can see what it can. In your
Sources list it carries a small **MEM** badge, a title like "What ChatGPT
remembers about you — 12 facts," a line saying where it came from, and
"answers only, not voice" — so this is never invisible. If your twin's tone
ever drifts, this import isn't why.

## Nothing syncs

This is a one-time paste, not a connection. There's no refresh button and no
"last synced" line, because nothing here stays live after you add it. If an
assistant learns something new about you next month, the only way to bring it
in is to paste again.

## Removing it

Delete it the same way as any other source, from Knowledge → Sources — there's
no separate control for a memory import. Deleting it takes the source and
everything indexed from it out together, at the next
[train](/docs/training) — nothing is kept behind a flag, and none of it lingers
past that rebuild. Trains run overnight on their own, so if you want it gone
now rather than tonight, delete it and then press **Train now**: until that
rebuild lands, your twin can still answer from it.

That only removes hiy's copy. It doesn't touch anything ChatGPT, Claude or
Gemini itself still holds — clearing it from the assistant's own memory means
using that assistant's own settings.