Import from ChatGPT and Claude
Bring what you have told ChatGPT and Claude into Notes from your own data export, or paste what your AI remembers, and review every suggestion before it reaches your vault.
Years of chats with ChatGPT or Claude hold a lot about you and your work: the projects you run, the tools you prefer, decisions you made and the people you work with. Notes can read that history and suggest notes from it.
Neither ChatGPT nor Claude lets another app read your chats, and Notes never signs in to them for you. So the import starts from the data export each app lets you download. Open Import in Notes (the upload button above the file tree) and find Bring in what your AI knows. There are three cards.
Import from ChatGPT
- In ChatGPT, open Settings → Data controls → Export data and confirm.
- OpenAI emails you a link, usually within a few hours. Download the zip.
- In Notes, choose Import from ChatGPT → Choose the export and pick the zip. You don't need to unzip it. If your
browser unzipped it for you (Safari does), pick the
conversations.jsonfile from the folder instead.
Import from Claude
- In Claude, open Settings → Privacy → Export data.
- Anthropic emails you a link. Download the zip.
- In Notes, choose Import from Claude → Choose the export and pick the zip, or the
conversations.jsonandprojects.jsonfiles from it.
Claude's export also has your projects. Each project becomes one item to import, made of its description, its instructions and its documents. Anthropic's example project is skipped.
You can also drop the zip onto either card.
Pick what Notes reads
Notes reads the export in your browser and lists every conversation with its date, number of messages and size. Then you choose:
- Search titles and From / To dates narrow the list.
- Length hides short conversations.
- Skip conversations that are mostly code hides chats where most of the text is code blocks.
- Select all, Select none and Select only the ones shown set the selection.
- Import into is your own vault. People who manage the org can choose the org vault instead, with a warning, because everyone in the org can read it.
Before you start, Notes shows what reading the selection will cost: a token count, and a dollar estimate when it knows the price of your org's summary model. It also says how many days that takes under your org's daily cap (Settings → How Notes learns).
The estimate counts about four characters per token, at most three chunks of 24,000 characters per conversation, the instructions and related notes sent with each chunk, and a typical answer. Long code blocks are shortened before the model reads them, so they cost little. The gateway records the real cost of each call.
What happens next
Notes sends the conversations you picked in batches. Each one is stored as a source, word for word, in the vault you chose. Then the loop reads them in the background, a few every 15 minutes, until the day's cap is spent. It carries on the next day. The first few start right away.
For each conversation, Notes looks for things that will still be true in months:
- your role, projects and goals, the tools you use, your preferences and how you like things done;
- decisions you made, and why;
- how-tos you worked out and will use again;
- the people and teams you work with;
- your own terms and names.
It skips small talk, one-off questions, drafts and code written for a single task. It counts what the assistant said only when you agreed with it or built on it. It links to your existing notes where they fit, and adds to them before it creates new ones.
Every suggestion waits in the Inbox. Imports never write to your vault on their own, not even your personal vault, where the other suggestions from Notes are added straight away. Accept, edit or reject each one, or use Apply all.
The progress view shows how many conversations were sent, queued, already in Notes, skipped or failed, and how many Notes has read so far.
Pause and resume
Pause stops after the current batch. If you close the tab, choose the same export again later: Notes recognises it and offers Resume, which sends only the conversations still left. It keeps your selection and progress in your browser's local storage, under a fingerprint of the export.
Importing again
A conversation already imported into the same vault is skipped, matched by its ChatGPT or Claude id or by identical text. Importing the same export twice adds nothing. A conversation you continued after an earlier import is skipped too, because it has the same id.
Paste what your AI remembers
This one is quicker. It is also the way to bring over ChatGPT's saved memories, which the export leaves out.
Choose Get the prompt and copy it. It reads:
List everything you know or remember about me: my work and role, current projects, tools and preferences, people I work with, decisions I've made, and how I like things done. Use short bullet points grouped by topic. Don't include anything sensitive like passwords or keys.Send it in a new chat in ChatGPT or Claude.
Paste the answer into the box and choose Send to Notes.
Notes reads it right away. The suggestions arrive in your Inbox within a minute or two and wait for you there. The answer can be up to 50,000 characters. Text that looks like a password or key is refused, so remove it first.
Privacy
- The export stays on your device. Notes opens the zip in your browser, reads only
conversations.jsonandprojects.jsonfrom it, and never uploads the zip. Images, audio and other files in the export are not read. - Only the conversations you pick are sent, and only to your own Venom Cloud organisation.
- Text that looks like an API key, a token or a private key is replaced with
[secret removed]before a conversation is stored. The progress view shows how many were removed. - Notes makes no other network calls for the import. Nothing goes to OpenAI or Anthropic, and the help links on the cards are plain links that load nothing until you click them.
- Imported conversations are sources like anything else you capture. People who can read the vault can read them: only you for your own vault, everyone in the org for the org vault. The audit log records how many were imported, never their titles or text.
Limits and troubleshooting
| What | Limit |
|---|---|
| One conversation | 1,000,000 characters. The model reads up to three chunks of 24,000 characters, after long code blocks are shortened |
| One request | 25 conversations, about 1 MB. The page batches for you |
| The export | Any size: it is read a piece at a time and never loaded whole |
- "has no conversations.json": you picked another zip. Pick the one from the email.
- "could not be read: it may be damaged or incomplete": the download stopped part way. Download the export again.
- Nothing to import: every conversation was empty or had nothing you wrote.
- Paused, "Lost the connection": check your connection, then choose Continue.
- Nothing in the Inbox yet: Notes reads a few conversations every 15 minutes, and stops for the day when the cap is spent. See the counts in the progress view, or raise the cap under How Notes learns.
The import routes are in the Venom Notes API. How the Inbox and the daily cap work is in How Notes learns.