AI-Hints is a powerful Anki add-on that uses Artificial Intelligence to generate helpful hints or multiple-choice options for your flashcards during review. It helps simulate real exam conditions by including the correct answer alongside plausible distractors even for open-ended cards.
Install from anki web
github:https://github.com/athulkrishna2015/AI-Hints
Features
- Multi-Provider Support: Supports OpenAI, Anthropic, Gemini, Groq, DeepSeek, NVIDIA, Mistral, Grok, OpenRouter, Together AI, Hugging Face, SambaNova, Cerebras, and any OpenAI-compatible local API (like Ollama or LM Studio).
- Automatic Fallback: If your primary AI provider fails (e.g., rate limits or API downtime), the add-on automatically attempts to generate hints using your next configured provider.
- Model Fallbacks: If a configured model is unavailable or returns unusable output, the add-on can try fallback models for the same provider before moving on.
- Current-Card Hints: Generated data is scoped to the current card, so cloze deletion and Image Occlusion siblings can each have their own hints/options.
- Alt-Click Reveal: Alt-click the current cloze deletion or Image Occlusion mask to reveal only that card’s hints/options. Ctrl-click and editable review fields are left alone.
- Speed Focus Mode Friendly: Clicking AI-Hints buttons or Alt-click revealing hints restarts the Speed Focus Mode timer when that add-on is installed.
- Live Log Viewer: Debug issues in real-time with a built-in log viewer in the config dialog. It features smart scrolling (only scrolls to bottom if you are already there), auto-refresh, and a “Live” status indicator.
- Improved UI Reset: Hints and options are now consistently cleared and reset whenever you move to a new card, ensuring a clean state for every review session.
- Custom AI Support: Define your own custom endpoints and headers.
- Universal Compatibility: Works with Basic, Cloze, Image Occlusion, and custom note types.
- Smart Shuffling: Options are shuffled every time you review the card to prevent pattern memorization.
- Storage Modes: Choose between visible HTML (visible on all devices) or invisible JSON (cleaner look, requires add-on to render).
- Configurable Options: Set exactly how many MCQ options the AI should generate, including the correct answer (default is 4).
- MathJax-Aware Rendering: AI-Hints prioritizes standard LaTeX/MathJax delimiters. It strictly uses
\( ... \)for inline math and\[ ... \]for block math, explicitly avoiding$ ... $or$$ ... $$to ensure compatibility with all Anki platforms. It preserves and repairs common LaTeX/MathJax output, including escaped JSON backslashes and bare variables such aslambda_L. - Field Customization: Specify exactly which fields to send to the AI for each note type. Optimized for Cloze cards by default.
- Persistent Storage: Generated hints are saved directly in your card’s fields (e.g., “Extras” or “Back”), so they work on AnkiMobile and AnkiDroid too.
- Manual Control: Generate, show, or regenerate hints with buttons on the card, the review bar, or both.
Configuration
Go to Tools → Add-ons → AI-Hints → Config to open the graphical configuration window.
Configuration Options
- General Tab: Select your
ai_provider, configure the total number of MCQ options (options_count), choose whether the Generate button appears on the review card, the review bar, or both, and choose thestorage_mode("json"for a hidden data block or"html"for a visible block). - AI Providers Tab: Enter your
api_keysfor supported providers, change model names, configure a local LLM endpoint (like Ollama), or use the Custom Providers section to add, edit, or remove your own API endpoints and custom headers. - Local Fallback: Enable “Use Local AI as fallback” if you want Ollama/LM Studio to be tried after cloud providers fail. Selecting
localas the active provider always uses the configured local endpoint. - Advanced Tab:
system_prompt: The base instructions for the AI.Note Type Fields: A graphical dropdown selector that allows you to specify exactly which fields should be sent to the AI for each of your specific note types.Raw JSON Editor: Advanced users can edit the full config directly; when the raw editor is enabled, it is saved as the source of truth.
Get Your API Keys
To use this add-on, you need an API key from one of the supported providers.
Free / Freemium Providers (No credit card required)
- Google Gemini: aistudio.google.com
- Groq: console.groq.com
- SambaNova: cloud.sambanova.ai/apis
- Cerebras: cloud.cerebras.ai
- Hugging Face: huggingface.co
- OpenRouter: openrouter.ai (Offers many $0/token models)
- Ollama (Local AI): ollama.com (No API key or internet required)
Paid Providers
- OpenAI: platform.openai.com
- Anthropic (Claude): console.anthropic.com
- DeepSeek: platform.deepseek.com
- Together AI: api.together.xyz (Requires initial deposit)
- Mistral: console.mistral.ai
- Grok (xAI): console.x.ai
- NVIDIA: build.nvidia.com
Example Config (Advanced Users):
You can also use the Raw JSON Editor in the Advanced tab if you prefer editing JSON directly:
{
"ai_provider": "openai",
"storage_mode": "json",
"options_count": 4,
"api_keys": {
"openai": "sk-...",
"groq": "gsk_...",
"sambanova": "",
"cerebras": ""
},
"models": {
"gemini": "gemini-3-flash-preview",
"groq": "llama-3.1-8b-instant",
"openrouter": "meta-llama/llama-3.1-8b-instruct"
},
"model_fallbacks": {
"gemini": ["gemini-3.1-flash-lite-preview"],
"groq": ["llama-3.3-70b-versatile"]
},
"local_endpoint": {
"enabled": false,
"base_url": "http://localhost:11434/v1",
"model": "llama3",
"api_key": ""
},
"note_type_fields": {
"Cloze": ["Text"],
"Basic": ["Front", "Back"]
},
"custom_providers": {
"my-local-llm": {
"url": "http://localhost:8080/v1/chat/completions",
"api_key": "optional",
"model": "mistral-7b",
"headers": {
"X-Custom-Header": "value"
}
}
}
}
Development
See DEVELOPMENT.md for build instructions and technical details.
License
MIT
Changelog
1.0.4
- Enhanced Logging: Logs now show the specific model name being queried for every request.
- Advanced Log Filtering: Added a real-time search box to the Logs tab to filter entries by keyword (e.g., model name, provider, or error).
- Bug Fixes: Improved handling of nested LaTeX delimiters and double-escaped backslashes in AI responses.
1.0.3
- Dynamic Model Discovery: Added “Fetch” and “Fetch All” buttons to retrieve the latest models directly from AI provider APIs.
- Customizable Fallback Priority: New drag-and-drop list to set the preferred order for provider fallbacks.
- UI Upgrade: Replaced model name text fields with editable dropdowns containing popular model suggestions.
- Smarter Defaults: Updated default models to
gemini-2.0-flash-exp(Gemini),llama-3.3-70b-versatile(Groq), andgoogle/gemini-2.0-flash-exp:free(OpenRouter).
1.0.2
- LaTeX Instruction Refinement: Updated system prompts to strictly require
\(and\[delimiters and explicitly forbid$signs. - MathJax-Aware Rendering: Improved preservation and repair of common LaTeX output patterns.
1.0.0
- Initial Release: Multi-provider support with automatic fallback and persistent hint storage.
- MathJax Support: Core logic for handling LaTeX/MathJax in AI-generated hints and options.
- Live Log Viewer: Real-time debugging interface in the configuration dialog.
