Smart Browser Search — AI card finder for Anki

Ask for a card in plain English. Get matching cards, a copy-paste Anki search string, and related concepts — powered entirely by local AI, right inside the Browser.

Type "that syndrome with hypertension and low potassium" → finds your Conn's / hyperaldosteronism cards, even if the card never says "low potassium."

Built for anyone with a large collection (medical board prep, language learning, law, etc.) who can't always remember the exact words on a card.

Features

• Natural-language search — describe a topic, symptom, or concept; the assistant turns it into a real Anki search and shows matching cards
• Medical synonym expansion — expands acronyms (MI → STEMI), brand ↔ generic (Lasix ↔ furosemide), eponyms, and classic findings automatically
• Copy-paste search string — every answer includes a Browser-ready Anki query with Copy and Run in Browser buttons
• Picture search — indexes card images via local OCR/captioning, ideal for Image Occlusion where the answer lives inside the image
• Clarifying questions — asks for clarification when your request is ambiguous; tap a quick-reply chip to continue
• Related concepts — suggests adjacent topics when nothing matches exactly
• 100% local & private — runs on Ollama or any OpenAI-compatible server; nothing leaves your machine, no account required
• Read-only by construction — cannot edit, add, delete, or reschedule a card; one-click safety proof in Settings
• No heavy dependencies — pure Python standard library; won't break when Anki updates

Requirements

• Anki 24.06 or newer (Qt6 desktop build — not the mobile apps)
• A local AI runtime running on your machine:
  • Ollama — recommended, free, one-click installer
  • Or any OpenAI-compatible server: LM Studio, llama.cpp, Jan, GPT4All
• Disk space for one model — as little as ~2 GB on the Light tier; CPU-only works (just slower)

Setup

1 — Install a local AI runtime

Ollama (recommended): download from ollama.com and run the installer. It runs quietly in the background; the add-on finds it automatically.

Using LM Studio / llama.cpp instead? Install it, load a model, start its server, then in Settings set the backend to OpenAI-compatible and paste your server's URL (e.g. http://localhost:1234/v1).

2 — Install this add-on

Tools → Add-ons → Get Add-ons…, enter the add-on code, click OK, restart Anki.

3 — Download a model

Open Tools → Smart Browser Search: Settings…, pick a tier under Recommended models that matches your hardware, click Use these models, then Download with Ollama.

Choosing a model

All tiers also use nomic-embed-text for semantic search and moondream / qwen2.5vl:7b for picture search — both optional and only downloaded if you enable those features.

Not sure? Start on Light — it works on any machine and you can switch up in Settings at any time.

Usage

1. Open the Browse window (Tools → Browse or press b)
2. The Smart AI Search panel appears on the right — toggle it with Ctrl+Shift+F (rebindable in Settings)
3. Type your question and press Enter

The assistant replies with a short answer, a copy-paste Anki search string with Copy and Run in Browser buttons, a list of matching cards, and a row of related-concept chips. Click any result to jump to that card in the Browse table.

Picture search: click the 🖼 button in the composer to attach an image (e.g. a screenshot of an ECG), or enable Index card images for picture search in Settings to make Image Occlusion card contents searchable by text.

Semantic search (opt-in): turn on Enable semantic search in Settings and click Build index for meaning-based matching that works even when card and query share no words. The first build runs in the background; subsequent runs update only changed cards.

Settings

Open via Tools → Smart Browser Search: Settings…, the ⚙ button in the panel, or the Config button in Tools → Add-ons.

Key options:

• Backend / Endpoint URL — choose Ollama or OpenAI-compatible and set the server address
• Recommended models — pick Light / Medium / High, apply, and download in one place
• Scope — limit search and indexing to specific decks
• Shortcut / dock side — rebind the toggle key, move the panel left or right
• Safety self-check — scans the add-on's own source for any card-modifying call

Troubleshooting

Is my collection safe?

Yes. This add-on is read-only by construction — it only calls Anki's search and note-reading APIs, and has no code path that writes anything. The semantic index lives in a separate database (user_files/index.db) that physically cannot touch collection.anki2.

Run Settings → Run read-only safety self-check to verify this yourself — it scans the add-on's own source and reports the result.

Privacy

Everything runs on your machine. Your cards, queries, and images are sent only to the local AI server you configure on localhost. There is no telemetry, no account, and no external network calls of any kind.

Support & source

Bugs and feature requests: https://github.com/Dogebooch/anki-smart-browser-search

Source, architecture notes, and build instructions are in the repository above.

License

MIT