1. Capture asks nothing
On a computer, “Write a note…” sits at the top of your notes, and Enter saves. On a phone, “Say it or write it” sits at the bottom: Return adds a line, and the arrow saves. There is no title field, no folder to pick and no tag to add, because choosing where something goes is the moment most notes die.
- Voice. Tap the mic and talk, in any of about 100 languages. The recording is written down by Whisper on the server in the language you spoke, and the audio is not kept. Ukrainian and Russian are both tried and the surer version wins.
- Recorded conversations. While recording, switch on Conversation; a conversation can run for up to two hours. After you stop, the note arrives with one line per turn (“Speaker 1: …”), and “Who is Speaker 1?” renames every line at once.
- Photos. From the camera or the gallery. Photos you pick together make one note, up to ten of them.
- Inside a lens. Anything you write inside a lens is filed there.
Notes are saved on your device first, so capture works without a connection; they sync when you are back online.
2. The agent’s read: one line, or nothing
A second or two after you save, the agent reads the note. It sees the note, a few of your recent notes, your lenses, what it has learned about the people and things you mention, and, if your calendar is connected, the events around the time you wrote it. Everything it reads is treated as data, never as instructions.
Then it may add one line under your words, at most 90 characters:
- A fact, such as who someone is or what the note connects to.
- A status: “due Friday, 9:30”. The model only reports what the note says (“next Friday”, “end of month”, “12.11”); code turns it into a date in your time zone.
- A guess you answer with Yes or No. Yes turns it into a settled fact. Every answer has Undo.
Most short, clear notes get nothing, and then nothing shows below them: no badge, no “processed”, no spinner. A line that only restates your note, invents a source, carries a web address or gets a date wrong is dropped by code before you see it.
The read can also notice things and propose them, each one waiting for your tap:
- Duplicates. “You already have this note, from today at 16:01. Keep one?”
- Adds to. A note that continues an older one: “Merge?”
- A pile. “3 notes in one. Split?” Each part becomes its own note, in your own words.
- A list. “4 items. Make a checklist?” Three or more to-dos on one subject: “Collect them into a checklist?”
- A rambling dictation. “Polish text” shows a readable version; only “Use this” changes the note, with Undo.
- A name it can’t place. When a name or term shows up in two or more notes and nothing explains it: “Who is Marcus?” At most two such questions a day, never about secrets.
Beside the line, the agent writes a short reading of the note, shown as its context. You can clear it with Forget.
3. Lenses and handles: structure grows by itself
When a note and at least two others share a subject that none of your lenses covers, the agent makes a lens at once, gives it a short key (WAT for Waterside) and tells you: “New lens: Waterside · Undo”. A new note about that subject is filed there; a note already in one of your lenses stays where it is.
Every note has a permanent number, like N-45, and a handle: its number in its lens (WAT-3), numbered oldest first, or the number itself when it isn’t in a lens. References are written with the permanent number, and every mention is drawn as the note’s current handle, so “see N-45” shows as WAT-3 once that note joins Waterside, and nothing goes stale. Old handles keep working.
You stay in charge: rename a lens, remove it (Undo restores it), or take a note out. The agent never files a note back into a lens you took it out of, and never makes a lens you removed again. You can also ask for a view yourself (“everything about the move”): on a computer you preview what matches and save it as a lens; on a phone the lens is made at once.
Three lenses are always there: Urgent (notes with a due date), Fading and Archive.
4. Catch up: the one place where the agent asks
Everything the agent wants from you waits in Catch up, so your notes never fill up with questions:
- Needs you: overdue dates, guesses, questions about names, proposals, the weekly review, a calendar write that failed, a conversation that couldn’t be written down.
- Suggestions: polishing a dictation, putting a timed note on your calendar, removing a lens that went quiet.
- Updates: “I made a lens: Waterside”, “Your conversation from 14:00 is written down”.
Every card says which notes it is about and why it is there. The answers the agent wants are yellow, and yellow appears nowhere else in Noteer. A note with something waiting shows one quiet pointer line; a guess on a note written in the last hour keeps its Yes and No right there. Dismissals hold on all your devices.
5. Day and Google Calendar
Day shows what is due and what slipped, and, with Google Calendar connected, your events. Connecting asks Google for permission to see your calendar list and events, and to add events. Events are read when shown, kept for two minutes and never copied into Noteer. Each calendar is Work, Personal or hidden; only Work and Personal are read.
A note written during a meeting gets that meeting as context. “Add to calendar” on a note with a due day opens a sheet with the title, day, time read from the text, length and calendar; the event is created only when you tap Add, and says “From Noteer · N-45” so you can find your way back.
6. The Map
The Map lays your lenses out over time: one row per lens, today on the right. To-dos are solid, done ones carry a check, information is hollow. Notes that mention each other are linked, each link with the agent’s reason (“Answers N-12”). The things the agent knows about (people, projects) sit above the map; tap one to light up its notes. On the phone the details open as a sheet; there is also a plain list view with the same content.
7. Chat that answers from your notes
Chat answers only from your notes and shows the notes it used as chips you can open. Ask about everything, about one note (every note has its own thread), about a lens, or about notes you pick. Answers come in the language you ask in. When you explain who someone is, the agent remembers it in “What I know”, which you can read, edit and clear in Settings. A question that has nothing to do with your notes gets an offer to save it as a thought, not an essay.
8. Fading and the weekly review
A note fades 14 days after the later of its last touch and its due date. Opening, editing, answering, pinning or being cited in chat count as a touch. Faded notes leave your list and wait in Fading. Once a week, on the day and time you choose, Catch up offers the review: tap the notes to keep, and the rest go to the archive. Pinned and done notes never fade.
Nothing in this is deletion. Only you delete a note, with Delete forever: it asks once, then the note, its history, the agent’s line and its proposals are removed, and every other device drops it too.
9. Phone, computer and offline
Noteer runs in the browser with an interface made for each screen: the phone gets tabs and cards, the computer gets a rail, panes and keyboard shortcuts (J and K to move, 1–9 to answer, ⌘K to search). One account syncs both. Changes land on the device first and wait in a queue; a quiet line says if something is waiting, failing or was refused, and refused edits are kept, never dropped. When the server has a newer version of the app, it reloads itself the next time you come back to it.
10. Secrets
Passwords, PINs, door codes, Wi-Fi keys, card numbers, IBANs, API keys and seed phrases are found by code, not by a model, before any text is sent to the AI. The AI sees a placeholder such as “[secret: password]”. Each secret is stored encrypted with a key of your own and shown to you as written, with a tint. When Noteer isn’t sure, it asks “Is this a password?” once. Polish is never offered on a note that holds a secret.
More on what is stored where: privacy policy. Questions: FAQ.