Score Answer AI for Anki

AI-powered semantic evaluation for Anki type: cards, with multilingual feedback, configurable prompts, and multi-provider support.

What This Add-on Does

• Evaluates typed answers semantically (not only exact text matching)
• Gives structured feedback:
  • score (0-10) when available
  • improvement tips
  • review suggestion (Again, Hard, Good, Easy)
• Runs analysis in background to keep review flow responsive
• Supports multiple LLM providers from one config screen
• Supports multilingual analysis and UI localization

This add-on is designed for Anki cards using typed answers ({{type:...}}).

Key Features

• Multi-provider support: OpenAI, Gemini, Claude, DeepSeek, Groq, OpenRouter
• Analysis language control: choose output language independently of your answer language
• Interface language auto mode: config UI can follow Anki UI language
• Custom prompt system:
  • optional custom system prompt
  • optional custom analysis prompt template with variables
  • reset to language defaults
  • copy defaults to clipboard
• Custom model IDs:
  • add your own model IDs in provider tabs
  • persist custom IDs in config
• OpenRouter resiliency:
  • recommended openrouter/free
  • fallback-aware connection test behavior
• Error-safe scoring:
  • provider errors no longer show fake 5/10
  • displays N/A when analysis is unavailable

Supported Languages

Analysis feedback and UI labels currently support:

• English
• French
• Spanish
• German
• Russian
• Japanese
• Chinese
• Korean

Installation

1. Open Anki.
2. Go to Tools -> Add-ons -> Install from file... (or install by AnkiWeb code if published).
3. Restart Anki.

Quick Start

1. Open Tools -> AI Multi-Provider Configuration.
2. Select provider and model.
3. Add your API key.
4. Click Test API Connection.
5. Select Analysis language.
6. Save and review your type: cards as usual.

Configuration Guide

General Settings

• AI Provider: active provider used for analysis
• Analysis language: language for AI feedback (tips) and prompt intent
• Enable AI analysis: global on/off
• Max tokens: max output size
• Temperature: response creativity/variance
• Show Anki compare: toggle native Anki comparison block
• Show code compare: toggle side-by-side extracted text comparison

Prompt Customization

• Use custom prompt template:
  • disabled: fields are read-only and show default placeholders
  • enabled: fields become editable
• Custom system prompt: optional replacement for default system message
• Custom analysis prompt template: supports:
  • {question}
  • {expected_answer}
  • {user_answer}
  • {language}
• Reset prompts to defaults: injects default prompts for selected analysis language
• Copy default prompts: copies language-specific defaults to clipboard

Provider Tabs

Each provider tab includes:

• API key field
• model selector
• editable model field
• Add model ID input/button to append custom model IDs

Custom model IDs are saved per provider in config and restored on restart.

Always test a model ID before using it in real reviews.
Some models may intermittently fail or be temporarily unavailable for unknown provider-side reasons.

OpenRouter Notes

OpenRouter availability can vary by model/provider at runtime.

Recommended defaults:

• openrouter/free for highest compatibility
• use specific :free variants only when needed
• always run Test API Connection after changing model ID

If selected model test fails, the add-on may successfully validate via openrouter/free fallback.

Scoring Behavior

• Normal case: shows score and review suggestion
• Provider/API/parsing failure: shows N/A (not a fake numeric score)
• Keeps failure details in feedback text for easier troubleshooting

Troubleshooting

"Connection error" or "Provider returned error"

• Verify API key and provider account status
• Try another model ID
• For OpenRouter, start with openrouter/free
• Reduce traffic / retry later (temporary provider saturation can happen)

Always getting English feedback

• Check Analysis language in config
• Ensure custom prompts do not force English
• Use reset prompt defaults for the selected analysis language

Model works in one provider but not another

• Model IDs are provider-specific
• Use Add model ID and test each ID explicitly

Privacy

• Card question/answer content is sent to selected AI provider for analysis
• The add-on stores temporary in-memory cache for active session flow
• No long-term local analytics storage is implemented by default
• Provider-side retention policies depend on the provider you choose

Compatibility

• Tested on modern Anki 25.x builds
• Uses Anki reviewer hooks for typed-answer comparison and rendering

Screenshots

Contributing

Issues and improvements are welcome.
When reporting bugs, include:

• Anki version
• add-on version/commit
• provider + model ID
• exact error text
• steps to reproduce