Notosaurus: snap a school lesson with your phone, review the cards, send them to Anki

Anki Add-ons
1

Hi everyone,

I’d like to share Notosaurus, an add-on I built for my kids, who are in middle school. You take a photo of a lesson with your phone (a notebook page, handwriting included, a textbook page or a PDF), an AI drafts the cards, you review and fix them, and they go straight into Anki.

  • AnkiWeb: add-on code 1956931509, or search “Notosaurus” on AnkiWeb
  • Source (AGPL-3.0): github.com/cchabanois/notosaurus
  • Documentation: cchabanois.github.io/notosaurus

Yes, it’s AI. The AI is the intern here: it does the typing, the cards come from your own lesson, and nothing reaches Anki before a human has checked it.

What it makes

  • Languages first: vocabulary with article, gender and plural; whole sentences; natural audio on the back, embedded in the deck (plays on AnkiDroid/AnkiMobile, offline); typed answers and spelling dictations.
  • Questions and answers, cloze, multiple choice, true/false.
  • Diagrams: labels hidden automatically, one card per label, without drawing the boxes.
  • Formulas in MathJax, geometry with exact SVG figures, pictures drawn by an image model.
  • Optional helps on the back (why the answer is right, a genuine way to remember it), and an explanation of any card while reviewing.

How it works

  • The add-on starts a small web server with Anki. The phone opens its page by scanning a QR code, on the same Wi-Fi; only paired devices get in, and the settings open on the computer only.
  • Cards go straight into the open profile, through a bridge built into the add-on, then sync. Sending again updates the notes instead of duplicating them (stable GUIDs); .apkg export is there too.
  • It ships its own note types, kept up to date when the add-on changes them.
  • You bring your own AI key: Gemini, Claude, GPT or OpenRouter. About 1 US cent per lesson with the recommended Gemini model. Photos go only to the service you choose; no tracking.
  • One Anki profile per child, each with their own lessons.

Feedback welcome, especially on Windows and macOS

I only use Linux. Installation is tested automatically on Windows and macOS, but nobody has used it there for real yet: if you try it, I’d love to hear how the first start went. Bugs and ideas are welcome here or on GitHub.