AnkiWeb addon 678719976
Adds a local AI panel in Anki's Browser that turns plain-English requests into Anki searches, expands synonyms, and finds matching cards privately.
Open on AnkiWeb GitHub Find alternatives
active
| Min Anki | Max Anki | Updated |
|---|---|---|
| 2.1.1 | – | 2026-06-27 |
| 2.1.2 | + | 2026-06-27 |
Loading…
Shows offline synonyms from your own collection on Chinese review cards as known/total counts on the front and a clickable synonyms panel on the back.
Turns Anki into a markdown notebook where you write papers, organize them in folders, and generate Basic, reversible, and cloze cards from plain-text lines.
Adds an in-card AI tutor for on-the-spot explanations, mnemonics, translations, image understanding, answer grading, and saved answers during review, using free or user-supplied models.
Provides a keyboard-first search palette for large medical collections with typo/alias matching, native Anki syntax, local semantic search, deck filtering, inline previews, editing, and safe bulk card actions.
Enables semantic and keyword search across selected Anki notes, generating cited AI answers through local or cloud providers.
Shows an instant popup dictionary when you select English words during Anki review, with Portuguese (BR) translation, definitions, phonetics, audio pronunciation, synonyms, and offline caching.
Fix 2.1.2 : Crashed during search item - Mostly on long cards
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.
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).
Tools → Add-ons → Get Add-ons…, enter the add-on code, click OK, restart Anki.
Open Tools → Smart Browser Search: Settings…, pick a tier under Recommended models that matches your hardware, click Use these models, then Download with Ollama.
| Tier | Chat model | Download | GPU memory | Good for |
|---|---|---|---|---|
| Light (default) | qwen2.5:3b-instruct | ~1.9 GB | ~4 GB or CPU | Any machine, laptops, integrated graphics |
| Medium | qwen2.5:7b-instruct | ~4.7 GB | ~8 GB | Mainstream gaming GPU (RTX 3060 / 4060 class) |
| High | qwen2.5:14b-instruct | ~9 GB | ~16 GB | High-end GPU (RTX 4070 Ti+, RX 7900 XT/XTX, M-series 32 GB+) |
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.
b)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.
Open via Tools → Smart Browser Search: Settings…, the ⚙ button in the panel, or the Config button in Tools → Add-ons.
Key options:
| Problem | Fix |
|---|---|
| Yellow "Local AI offline" banner | Start Ollama / your server, then click Retry in the banner |
| "Model not installed" banner | Settings → Download with Ollama, or run the ollama pull … shown |
| Test connection fails | Check the Endpoint URL — Ollama: http://localhost:11434; OpenAI-compatible: must end in /v1 |
| Slow first query | Model is cold-loading into memory; later queries are fast |
| Weak results | Switch to a heavier tier, or enable semantic search and build an index |
| Semantic search finds nothing | Enable it in Settings and click Build / update index |
| Panel doesn't appear | Confirm you're in Browse, then press Ctrl+Shift+F |
| Download with Ollama grayed out | That button is Ollama-only; load models in LM Studio / llama.cpp and type their names into the Models fields |
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.
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.
Bugs and feature requests: https://github.com/Dogebooch/anki-smart-browser-search
Source, architecture notes, and build instructions are in the repository above.
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.
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).
Tools → Add-ons → Get Add-ons…, enter the add-on code, click OK, restart Anki.
Open Tools → Smart Browser Search: Settings…, pick a tier under Recommended models that matches your hardware, click Use these models, then Download with Ollama.
| Tier | Chat model | Download | GPU memory | Good for |
|---|---|---|---|---|
| Light (default) | qwen2.5:3b-instruct |
~1.9 GB | ~4 GB or CPU | Any machine, laptops, integrated graphics |
| Medium | qwen2.5:7b-instruct |
~4.7 GB | ~8 GB | Mainstream gaming GPU (RTX 3060 / 4060 class) |
| High | qwen2.5:14b-instruct |
~9 GB | ~16 GB | High-end GPU (RTX 4070 Ti+, RX 7900 XT/XTX, M-series 32 GB+) |
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.
b)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.
Open via Tools → Smart Browser Search: Settings…, the ⚙ button in the panel, or the Config button in Tools → Add-ons.
Key options:
| Problem | Fix |
|---|---|
| Yellow "Local AI offline" banner | Start Ollama / your server, then click Retry in the banner |
| "Model not installed" banner | Settings → Download with Ollama, or run the ollama pull … shown |
| Test connection fails | Check the Endpoint URL — Ollama: http://localhost:11434; OpenAI-compatible: must end in /v1 |
| Slow first query | Model is cold-loading into memory; later queries are fast |
| Weak results | Switch to a heavier tier, or enable semantic search and build an index |
| Semantic search finds nothing | Enable it in Settings and click Build / update index |
| Panel doesn't appear | Confirm you're in Browse, then press Ctrl+Shift+F |
| Download with Ollama grayed out | That button is Ollama-only; load models in LM Studio / llama.cpp and type their names into the Models fields |
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.
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.
Bugs and feature requests: https://github.com/Dogebooch/anki-smart-browser-search
Source, architecture notes, and build instructions are in the repository above.