Sometimes the thing you want to remember isn't a thought — it's a document. Meeting notes someone sent you, a contract you want to be able to recall the terms of, a long-form article you'll want to come back to. Attach it to a message and Working Memory reads it into your memory.
How to attach a file
In the composer, tap the attach button next to the mic. Pick a PDF or a text file from your device. You can add a short note alongside it if you want (e.g. "Q3 board pack — read before friday") and the AI will associate the two.
How long it takes
Most files are processed instantly:
- Anything under about 100KB — short PDFs, plain-text notes, short transcripts — gets read straight away.
- Larger files (multi-page PDFs, long transcripts, long articles) take a bit longer. You'll see a placeholder on the timeline that resolves when the AI is done.
In both cases the result is the same: the document text is added to your searchable memory, and any structure the AI noticed — people mentioned, dates, action items, decisions — is pulled out and tagged.
Asking about a document later
Once it's in, you don't need to remember where you put it. Just ask:
- "what was the cancellation clause in the gym contract?"
- "who was on the action-item list from monday's planning doc?"
- "what did the Q3 board pack say about hiring?"
The AI looks across everything you've imported and answers from the source. See Asking questions for more on how recall works.
Good fits for imports
Anything you'd want to come back to without re-reading the whole thing:
- Meeting notes — your own or someone else's.
- Contracts and agreements — terms, dates, parties.
- Long-form pieces — research notes, essays, transcripts of talks.
- Reference material — the part of a manual you actually use.
What doesn't work
- Images. No OCR or vision yet. If the document is a photo of text, the AI won't read it.
- Audio files. Don't upload an MP3 — use a voice message instead. See Voice messages.
- Spreadsheets. Export to CSV or plain text first; Excel and Numbers files aren't read directly.
If you've got something that doesn't fit those categories and you'd like it to, let us know — it helps us prioritise what to add next.