Capture is half the loop. The other half is getting things back. Whenever you want to know something you've already told Working Memory, type or speak a question in the composer, and the AI pulls the relevant bits from your history.

How to ask

Ask the way you'd ask a friend who's been listening:

You don't need a special prefix, command, or syntax. The AI works out from the phrasing that you're asking rather than capturing. Questions tend to start with a question word ("what", "when", "who", "how") or be phrased as a request ("remind me what…", "tell me about…", "find…") — but if it's ambiguous, the AI errs on the side of asking, not adding a new note.

How retrieval works, in plain English

The AI looks across everything you've ever told it and pulls the bits that match your question, in two ways at once:

The result is a short, focused answer drawn from your actual notes — with the original context attached so you can see where it came from.

Asking the same thing different ways

You don't have to remember which words you used when you first said it. All of these will find the same note:

Phrase it however it comes to you. The AI does the bridging.

Asking about time

Time-scoped questions work too:

Your timezone is used to resolve "yesterday", "last week", and similar. See Timezones for how that's set.

When you ask something you never told it

If you ask about something you haven't actually said before, the AI says so honestly — it won't make up an answer to look helpful. That's deliberate: a memory you can't trust is worse than no memory. If you get "I don't see anything about that", the answer is to add the note now.

Where to go next