AA Notes Canvas

Inside the project

Prepared with the project’s development agent from source and retained evidence. This is a project-owner assessment, not an independent customer review. Dated checks were not rerun for this website.

The Notes module is the part of AssistantApp a visitor would recognise first, and the part that has quietly become its workbench. It started as a plain note panel and now has three layers. The first is a word processor: a TextKit-based editor with a nine-tab ribbon, Word built-in styles, comments, footnotes, fields, mail merge, outline, page layout and print preview, opening and saving Word, OpenDocument, RTF, HTML and plain text entirely offline, with its own PDF and EPUB writers and a hand-rolled ZIP so .docx and .epub need no third-party library. The second is a reading layer: Markdown and LaTeX typeset on the device by an authored maths engine, with Edit, Split, Reading, Canvas and PDF-map modes, syntax-coloured code blocks and a repair pass for pasted text that arrived in the wrong encoding. The third, and the one the owner spent most of August and September on, is the Canvas. A note can switch into an infinite spatial surface where nineteen kinds of frame coexist: text written straight onto the surface, shapes, images, charts, drawings, code boxes that check themselves against a real Xcode project, web pages, YouTube, Discord channels, social cards, voice memos with transcripts, music scores, a DICOM viewer frame hosting the app's existing imaging code, a developer command console, an algebra trainer, a spatial study and, most recently, Live Instruments. Frames drag, resize, rotate, snap to guides, group, align and distribute; the canvas has a minimap, deep search that matches meaning as well as words, a presentation deck, PDF export and a shared transport so padlocked audio clips play on one clock. Live Instruments are the research edge: an AI proposes a small interactive instrument as data, the native app validates it, computes what it may see and do, renders it in a six-band frame, and only the user may keep a finding or append it to the note. The first instrument, an Inference Route Explorer, compares the app's AI routes by declared fit and by a bounded 48-token speed probe, and labels every value with its basis so a preference is never ranked as a measurement. All of this is built for macOS only, is uneven in how far it has been exercised at runtime, and has no public release. New native captures from a synthetic demo note now accompany the dated capability review below.

The 18 October 2026 screenshot shows the actual Notes Canvas with its built-in Inference Route Explorer, added to a newly created empty demo note while the private library was hidden. The instrument compares declared route capabilities and exposes evidence labels, a proposed policy and a Keep finding action. The displayed 97.39 chars/s value is a retained earlier probe record; no probe ran during this capture, and it is not a new timing benchmark. This verifies the visible interface for one bounded workflow, not every document, canvas or export feature.

Features & evidence

Workflow walkthrough

A guide to the native workflow. These steps describe the app; only the recordings and images embedded on this page are available as website demonstrations.

1. Open a note in the Notes panel and switch the render mode picker to Canvas. The note body stays available in Edit or Split; the canvas is a dot-grid surface with an add-frame toolbar.
2. Click anywhere and start typing: a bare caret becomes a text frame. Paste a picture onto the text and the text flows around it; drag a social chip from the palette to drop a link card.
3. Add a Module frame and choose Response Curve. Drag points on the curve, then flip the frame to Code and see the same model as JSON; edit a value and apply it back.
4. Insert an Instrument: Inference Route Explorer. Move the Privacy control to local-first and read the consequence line; the cloud routes drop out and every remaining value shows its basis chip (asserted, measured, preference).
5. Press Run probe. A single 48-token timing appears labelled 'one sample, shown not ranked'. Pin a finding; it stays attached with a staleness mark if the table changes.
6. Choose 'Add finding to note'. The approval sheet shows what will be appended and the prior state; confirm, and the sentence lands in the note body with a revert affordance. Export the note as PDF from the file menu.
7. Status: proposed capture that does not exist. Nothing here is playable on the website and there is no existing native capture; the steps are reconstructed from source and this session's records, not from a run.