HandNote

Handwritten notebook pages for Obsidian, made for the Apple Pencil and working with touch and mouse. Write with pressure-sensitive pens, select with a lasso, and convert handwriting into editable text — on your device, without an account or API key. Formulas are edited as LaTeX.

中文说明

HandNote is an independent, third-party community plugin. It is not made by, affiliated with or endorsed by Obsidian, MyScript, Goodnotes, Xmind or Excalidraw.

Features

• Notebook pages in .handnote files: A4 / A5 / A3 / Letter, blank, ruled, grid or dotted paper, many pages per note, a page sidebar to reorder pages. Embed a note in Markdown with ![[My note.handnote]].
• Pens: fountain pen with real Apple Pencil pressure, pencil, highlighter, dashed and dotted strokes, shape pen (lines, arrows, circles, rectangles are recognised as you draw), laser pointer.
• Eraser: precise, partial or whole-stroke; highlighter-only.
• Objects: text boxes, images (vault attachments or embedded), shapes and connectors, editable formulas, mind maps.
• Lasso: free-form or rectangular. Move, resize, rotate, copy, cut, recolour, group. A loop around part of a text picks just those characters.
• Convert handwriting (lasso → Convert):
  • Text (Chinese and English), in place, one undo step, with "Restore ink" and candidates for uncertain characters. Recognition runs on your device after a one-time model download.
  • Math: this version has no on-device math recognition. Convert → Math lets you type the formula as LaTeX (with a live preview; the ink is kept or replaced, as you choose), or recognise it with MyScript (your own keys).
  • Formulas can be edited, moved, resized, copied and copied as Markdown ($$…$$).
• AI tools (optional, your own provider key): turn a sketch into a mind map or a flowchart, read text from a picture.
• Safe storage: every save checks that the file was not changed by sync first; a recovery journal and backups protect unsaved strokes; damaged or newer-format notes open read-only and are never overwritten.

Platforms

Platform	Status
iPad (Obsidian for iPadOS) with Apple Pencil	Primary target. Pressure, 240 Hz Pencil input, palm rejection, finger pan / pinch, two-finger undo.
Windows / macOS / Linux desktop	Mouse, pen tablet, touch screen, keyboard shortcuts.
iPhone / Android	Should work (no Node or Electron APIs are used), but is not tested.

Minimum Obsidian version: 1.11.4 (secret storage for the optional API keys).

Writing

Action	Pencil / pen	Finger	Mouse / keyboard
Write	draw	— (unless "finger drawing" is on)	drag
Pan / zoom	—	one-finger drag / pinch	space + drag, Ctrl + wheel
Undo / redo	—	two- / three-finger tap	Ctrl+Z / Ctrl+Shift+Z
Edit a text	tap it (lasso or text tool)	tap it	click it (lasso or text tool)
Pen ⇄ eraser	—	double-tap with one finger	1 / E

Lasso and conversion

1. Pick the lasso and draw a loop around handwriting. The loop stays as the outline of the selection.
2. Tap Convert → Text. The handwriting is replaced in place by editable text, about as large as you wrote it. Convert → Math opens the LaTeX editor (or uses MyScript, if you set it up).
3. The result stays selected: drag it, tap it to type, pick another candidate in the bar below it, or press Restore ink.

On-device recognition and model downloads

Settings → HandNote → Local recognition. The recognition runtime (ONNX Runtime Web 1.24.2, MIT) is built into the plugin; it is never downloaded and starts only when you convert. The recognition models are not built in: each one is downloaded separately, and only when you ask.

Pack	Download size	Downloaded from (first mirror that answers)	Used for	Licence of the files
Chinese text — PaddleOCR PP-OCRv5 mobile rec (ONNX) + dictionary	15.8 MB	PaddleOCR's official download (Baidu, paddle-model-ecology.bj.bcebos.com), Hugging Face PaddlePaddle/PP-OCRv5_mobile_rec_onnx (or hf-mirror.com); dictionary from the PaddleOCR GitHub repository (or its jsDelivr copy)	Chinese and mixed lines	Apache-2.0 (model card)
English text — PaddleOCR en_PP-OCRv5 mobile rec (ONNX) + dictionary	7.5 MB	as above (en_PP-OCRv5_mobile_rec_onnx)	English lines	Apache-2.0 (model card)
Hand font — Excalifont + Xiaolai (optional)	up to 12.1 MB, loaded per character	Excalidraw's npm package / repository (npmmirror, jsDelivr, GitHub, Fastly)	hand-lettered look for converted text	SIL OFL 1.1

• When: only after you press a pack's Download button in Local recognition, or accept the offer shown when you convert without the model. The size is shown first. Nothing is downloaded at startup, on install, on update or in the background.
• Checked: every file has a fixed size and SHA-256 in the plugin; a file that does not match is discarded. There is no automatic model update: new models only come with a new plugin version, and you still press the button.
• Stored: in this device's app storage (IndexedDB, per device). Never in your vault, never synced. Uninstalling HandNote does not delete them; delete them in Local recognition first.
• Uploaded: nothing. The download requests are plain file requests (short HEAD requests first to find a mirror that answers). Your notes and your handwriting never leave the device for local recognition.
• Offline: after the download, recognition works without a connection. Without the models, writing and everything else still works; Convert → Text can use the built-in offline Chinese engine.
• Blocked mirrors: Import files accepts the same files from disk (same hash check). Run self-test shows what works on this device and which mirrors answer.
• Local recognition needs iPadOS / iOS 16.4 or newer on mobile devices (older systems can use MyScript).

Optional cloud services

Nothing is sent anywhere unless you set it up yourself:

• MyScript (Settings → HandNote → Recognition): enter your own MyScript keys. The strokes of a selection are sent to MyScript after a confirmation. MyScript's terms and pricing apply.
• AI providers (Settings → HandNote → AI): OpenAI, Anthropic, Google Gemini, DeepSeek, Alibaba Cloud Qwen, or any OpenAI-compatible endpoint, with your own key. A picture of the selection is sent after a confirmation. The provider's pricing applies.

Keys are kept in Obsidian's secret storage, never in settings files or notes.

Privacy

No telemetry, no analytics, no account, no server of its own. The full list of network requests — service, purpose, data sent, cost, offline behaviour — is in PRIVACY.md.

Installation

From Obsidian: Settings → Community plugins → Browse → search "HandNote" → Install → Enable.

Manual: download main.js, manifest.json and styles.css from the latest release into <vault>/.obsidian/plugins/handnote/, then enable HandNote in Settings → Community plugins. Each release also has SHA256SUMS.txt to verify the files.

FAQ

See FAQ.md. Short answers:

• Where are my notes? In .handnote files in your vault, synced by whatever you use (Obsidian Sync, iCloud, Remotely Save…).
• The model download hangs. Run the self-test to see which mirrors your network reaches. Use "Import files" if none does.
• English recognition is weak. Install the English pack. Lines written in Latin script use it automatically.

Known limitations

• No on-device math recognition. No handwritten-math model with clear licence terms is included yet. Type formulas as LaTeX, or use MyScript with your own keys. Notes with formulas made by any HandNote version open and edit normally.
• Handwriting recognition is trained on general handwriting: unusual handwriting, very small writing or mixed scripts in one line are read less reliably. Uncertain results are shown with candidates, and the ink is kept until you choose.
• A note made with a newer HandNote opens read-only in an older one. Update HandNote on every device that edits the same notes.
• Phone layouts are not optimised.
• main.js is about 5.9 MB because the recognition runtime is built in. Obsidian Sync's Standard plan does not sync files over 5 MB, so with that plan install HandNote on each device from Community plugins (your notes sync as usual).

Feedback

Bugs and ideas: issues (please do not attach private notes or keys). Security problems: see SECURITY.md.

Licence

HandNote's own code: MIT. Parts made by others keep their own licences and are not covered by HandNote's licence: ONNX Runtime Web inside main.js (MIT, © Microsoft Corporation), the Chinese stroke data inside main.js (Arphic Public License, vendor/hanzi-mmah/), and the optional downloads (Apache-2.0, SIL OFL 1.1). Details and licence texts: THIRD_PARTY_NOTICES.md.