Generate vocabulary cards

This recipe creates contextual Japanese vocabulary cards for 会議 and adds only the candidates you approve to Japanese::Vocabulary.

Create a workspace prompt

From the directory where you keep deck configuration and prompts, run:

anki-llm workspace init
anki-llm generate-init

Choose Japanese::Vocabulary and the intended note type. The wizard samples existing notes and generates a prompt with a matching field_map, output shape, and style. Review the saved file in prompts/ and refine its instructions or one-shot JSON example.

For a prompt you can inspect immediately, see the vocabulary example. The Japanese contextual example shows a richer card and post-selection processing.

Tip

The prompt’s example is the strongest signal for field formatting. Make it a realistic card with exactly the keys declared in field_map.

Launch the generator

Start the TUI from the workspace:

anki-llm generate

A single prompt is selected automatically. With multiple prompts, choose one in the picker. Press Ctrl+O whenever you want to switch the model used by later requests.

See Generate cards for prompt discovery, model selection, thinking display, and the complete review workflow.

Enter one or more terms

Type 会議 and press Enter. To generate several terms in one session, press Tab after each queued term and Enter on the last one. You can also paste newline-separated terms.

The model returns several contextual candidates. Session token use and estimated cost appear in the sidebar while requests run.

Review every candidate

On the selection screen, move through the candidate list and inspect every mapped field. Check that:

• the sentence uses 会議 naturally and reflects the intended meaning;
• the English translation matches the Japanese;
• readings and furigana use your deck’s conventions;
• explanations add useful nuance instead of repeating the definition;
• HTML matches the card template.

Use Space to select a card, e to edit it in $EDITOR, and R to regenerate it with feedback. Use r for more candidates for 会議, or t to generate another term.

Exact matches on the note type’s first field are marked [dup] and shown as a diff against the existing Anki note. Leave them unselected unless the new card is intentional. Pressing f force-selects a duplicate.

Caution

Duplicate detection is an exact first-field comparison. It cannot replace your review for near-duplicates or semantically equivalent cards.

Use quality processing when needed

A generation prompt can run transforms and checks before or after selection. For example, use a transform to normalize furigana and a post-selection check to evaluate whether the Japanese sounds natural.

Post-selection work runs only for accepted cards, which keeps expensive checks focused. A flag opens a review path, while a reject removes the card. See Processing steps for the configuration and verdict rules.

Preview audio and import

If the prompt contains a tts: block, press p to hear the focused card before selection. Audio for selected cards is finalized when they are imported.

Press Enter after selecting the cards you want. Confirm the final review to add them to Anki. From the summary, press p to replay imported audio, n for another term, r to generate the last term again, or q to quit.

For an extra review boundary, generate to a local file instead of importing directly:

anki-llm generate "会議" --output meeting-cards.yaml

Review the file, then import it:

anki-llm import meeting-cards.yaml --deck "Japanese::Vocabulary"

The generate command reference lists output, count, retry, model, copy, logging, and dry-run options without duplicating them here.