VocaBuilder

A Raycast extension that translates words and short text between languages with the AI provider of your choice — Gemini, Anthropic, OpenAI, or any OpenAI-compatible server — and saves a local translation history to help you build vocabulary over time.

Features

Bring your own AI provider — use a Gemini, Anthropic, or OpenAI API key, or point the extension at an OpenAI-compatible server such as OpenRouter, Ollama, or LM Studio
Multi-sense word translations — for ambiguous words, review up to 5 senses, each with part of speech, pronunciation transcription, and example sentences, and save the one you want to keep
Typo correction — misspelled word input is auto-corrected before translating, with a visual indicator showing the original input
Word pronunciation — hear the source word or its translation through Gemini or OpenAI speech, or the macOS system voice for providers without a speech model
Flashcards — review saved word translations with spaced repetition
Separate history per language pair — 17 languages in any pairing, switchable from a dropdown inside each command; each pair keeps its own history and flashcard deck
History export — export saved history as JSON, an Anki-ready Basic note file, or Quizlet-ready TSV
Clipboard suggestion — optionally prefill a safe word or short phrase from the clipboard when the command opens

Getting Started

Get an API key from Google AI Studio (Gemini, free tier available), the Anthropic Console, or the OpenAI Platform.
Run the Translate command and paste the key when prompted. With Provider left on Auto-Detect, the extension recognizes Gemini (AIza…), Anthropic (sk-ant-…), and OpenAI (sk-…) keys.
For an OpenAI-compatible server, open the extension preferences, set Provider to OpenAI-Compatible Server, and fill in API Server URL (for example https://openrouter.ai/api/v1 or http://localhost:11434/v1) and Translation Model. Local servers that need no key accept any value in API Key.
(Optional) In the preferences, set default languages, override the translation or speech model, choose the macOS system voice for pronunciation, or enable clipboard suggestions on open.

Provider	Default translation model	Pronunciation
Gemini	`gemini-3.5-flash-lite`	`gemini-3.1-flash-tts-preview`
Anthropic	`claude-haiku-5-5`	macOS system voice
OpenAI	`gpt-6-luna`	`gpt-4o-mini-tts`
OpenAI-compatible	none — set Translation Model	macOS system voice

OpenAI speech overrides tts-1 and tts-1-hd use the macOS system voice to preserve the selected pronunciation language. Pronunciation is unavailable when that language has no system voice.

API Server URL also routes Gemini, Anthropic, or OpenAI traffic through a proxy or gateway. Your API key is sent to whichever host you configure, so plain http is accepted only for local addresses, and URLs carrying credentials or a query string are rejected.

Commands

Command	Description
Translate	Translate a word or short text
Translation History	Open your saved translations
View Flashcards	Review words you've saved

Supported Languages

English, Ukrainian, Polish, German, French, Spanish, Italian, Portuguese, Dutch, Czech, Swedish, Japanese, Korean, Chinese, Turkish, Russian, Belarusian

Keyboard Shortcuts

Shortcut	Action
`⌘ C`	Copy translation
`⌘ ⇧ L`	Pronounce source word
`⌘ ⇧ R`	Pronounce translation
`⌘ 1` → `⌘ 5`	Pick a word sense and save it to history
`⌘ ⇧ H`	Open History (from Translate)
`⌘ ⇧ T`	Toggle source and target languages
`⌘ D`	Delete entry (from History)
`⌘ ⇧ D`	Clear all history
`⌘ E`	Export history as JSON
`⌘ ⇧ A`	Export word history for Anki
`⌘ ⇧ Q`	Export word history for Quizlet

Development

Use the Node version pinned in .nvmrc — Node 22.22.2, which bundles npm 10.9.7 and matches the current Raycast extension toolchain. Then run the extension in Raycast dev mode:

nvm use        # Node 22.22.2 / npm 10.9.7
npm install
npm run dev

Regenerate package-lock.json only on npm 10.x. npm 11 records optional peer dependencies differently, producing a lockfile that passes locally but fails the store CI's npm ci.

Built with the Raycast API, TypeScript + React, and Zod for runtime validation. Each provider is a small adapter over plain fetch in src/lib/providers/; the prompt, response validation, and sense de-duplication are shared in src/lib/translation.ts.

Tests run with npm test. The translation quality evals read their keys from .env — copy .env.example to start. npm run eval scores the Gemini provider and needs only GEMINI_API_KEY, which the eval judge also uses; npm run eval:all scores Gemini, Anthropic, and OpenAI side by side and also needs ANTHROPIC_API_KEY and OPENAI_API_KEY.

License

MIT

Versions4

More AI providers, clearer word details, and Anki export - October 19, 2026

Translation now works with Gemini, Anthropic, OpenAI, or any server that speaks the OpenAI Chat Completions API, such as OpenRouter, Ollama, or LM Studio. With the new Provider preference left on Auto-Detect, the provider is recognized from your API key.
The Gemini API Key preference is now API Key, so Raycast asks for your key once after this update. Paste the same Gemini key to keep using Gemini.
Pronunciation uses OpenAI speech with an OpenAI key, Gemini speech with a Gemini key, and the macOS system voice for Anthropic and OpenAI-compatible servers. OpenAI's legacy tts-1 and tts-1-hd models also use the system voice to preserve the selected language. A new Pronunciation preference switches every provider to the macOS system voice. Starting any word or translation pronunciation, or leaving its view, stops speech already playing.
A word with more than five senses now shows the first five instead of failing the lookup.
Word detail panes now lead with the translation itself, set as the largest text in the pane, above a single dictionary line carrying the word you looked up, its pronunciation transcription, part of speech, key inflected forms, and a usage label such as formal or literary where one applies. Each element appears only when it exists for that word, so multi-word expressions and languages without a transcription convention simply show less.
Translations now request pronunciation transcriptions, inflected forms, and usage labels per sense, so a word that differs by part of speech — English record as a noun and as a verb — shows the transcription and forms belonging to the sense you picked.
Example sentences in the translate, history, and flashcard detail panes now show the source-language sentence first, with the target-language rendering beneath it. Empty stored examples are omitted without leaving formatting markers.
The looked-up word is now shown in bold inside the source-language example sentence, including inflected forms and mixed-script words such as Japanese Tシャツ in sentences without spaces. Han/kana lookups highlight only their exact text, even beside digits or Latin letters. Highlighting in spaced scripts requires a word boundary before the lookup, so unrelated fragments such as cat inside scatter stay plain.
Detail panes now list the pronunciation actions and their shortcuts beside the entry, and leave the row out for a language that neither your provider nor the macOS system voice can speak.
Pronunciation moved off ⌘O and ⌘⇧O onto ⌘⇧L for the source word and ⌘⇧R for the translation, in the translate, history, and flashcard views — the old chords are the ones Raycast reserves for Open and Open With.
Anki export now writes a two-column Front / Back file tagged as Anki's Basic note type, with the translation, its part of speech, and both example sentences formatted on the back of the card. Earlier exports used five separate columns, so a note type built from one of those will not line up — import into a fresh deck or remap the fields once.
Simplified translation and text-to-speech model preferences to accept any model ID your provider offers, without relying on hardcoded dropdown options. Leave the translation model empty to use Gemini 3.5 Flash-Lite, Claude Haiku 5.5, or GPT-6 Luna for your provider; leave the speech model empty for the provider's speech default.
Removed the separate reasoning preference so reasoning behavior follows the configured model.
Added an API Server URL preference so translation and pronunciation can be routed through a proxy or gateway, or to an OpenAI-compatible server, where it is required along with Translation Model. Enter a server root or a full API base such as https://openrouter.ai/api/v1. Your API key is sent to the host you configure, so plain http is accepted only for local addresses and credential-bearing URLs are rejected.
API requests reject redirects so your API key stays with the server you configured.
Reworked the translate screen while a translation is in progress: the word you are translating now appears as a row in the Translation section where its result will land, and your Recent translations stay on screen instead of the list clearing to an empty placeholder.
Word input now auto-translates after 0.7s instead of 1.5s.
Returning to a word already looked up in the same Translate session now restores its senses immediately, including after undoing a typo, without another API request. Cached results are kept separate for each language pair, provider, model, and API server.
Clipboard suggestions normalize extra spaces in short phrases before lookup and saving.
Failed exports and failed pronunciations now explain what went wrong instead of showing the raw system error and your file paths.

Configurable models and reasoning - June 8, 2026

Added a Reasoning Level preference for translations, defaulting to Minimum for the fastest Gemini responses.
Replaced the free-form translation model default with a curated Gemini model dropdown plus an advanced custom model override.
Replaced the free-form text-to-speech model default with a curated Gemini TTS dropdown plus an advanced custom model override.
Refreshed the extension icon.
Tightened the pronunciation prompt for more accurate text-to-speech output.

Switch language pairs, pronunciation, and configurable models - June 1, 2026

Added an in-command language pair selector: switch the source and target language straight from a dropdown in the Translate, Translation History, and View Flashcards commands, without opening preferences. The selection persists across commands.
Renamed the Source Language / Target Language preferences to Default Source Language / Default Target Language — they now apply until you pick a pair from the dropdown.
Added Translation Model and Text-to-Speech Model preferences so you can point to a newer Gemini model if a default is deprecated.
Reworked error handling into clearer, more consistent toasts, including deprecation-aware messages that tell you when a model is no longer available and how to switch.
Added a target-language purity rule so translations stay in the chosen language, reducing cases where words from another language leak into the output.
Reject non-existing/gibberish words instead of hallucinating translations.
Require example sentences to contain the original word, not a synonym.
Add post-validation that filters out senses with missing words in examples.
Handle phrasal verbs and idioms (e.g. "give up", "break the ice") as single vocabulary items.
Added word pronunciation via Gemini TTS (⌘O for source word, ⌘⇧O for translation).

Initial Release - April 8, 2026

Added Translate command to translate words and short texts between languages using Gemini.
Added Translation History command to browse saved translations.
Added Flashcards command for spaced-repetition review of saved words (SM-2 algorithm).
Added configurable source and target language support (17 languages).
Added color-coded part-of-speech chips: nouns (blue), verbs (red), adjectives (green), adverbs (magenta), prepositions (yellow), pronouns (purple), conjunctions (orange).
Added typo correction — misspelled words are auto-corrected before translation.
Added short text and sentence translation support with detail panel.
Added extension preferences for Gemini API Key, source/target language, and safe clipboard prefill behavior.
Added auto-translate with 1.5s debounce for word input, manual submit for text.
Added abort of in-flight requests on search text change.

Compatibility

macOS

VocaBuilder

Features

Bring your own AI provider — use a Gemini, Anthropic, or OpenAI API key, or point the extension at an OpenAI-compatible server such as OpenRouter, Ollama, or LM Studio
Multi-sense word translations — for ambiguous words, review up to 5 senses, each with part of speech, pronunciation transcription, and example sentences, and save the one you want to keep
Typo correction — misspelled word input is auto-corrected before translating, with a visual indicator showing the original input
Word pronunciation — hear the source word or its translation through Gemini or OpenAI speech, or the macOS system voice for providers without a speech model
Flashcards — review saved word translations with spaced repetition
Separate history per language pair — 17 languages in any pairing, switchable from a dropdown inside each command; each pair keeps its own history and flashcard deck
History export — export saved history as JSON, an Anki-ready Basic note file, or Quizlet-ready TSV
Clipboard suggestion — optionally prefill a safe word or short phrase from the clipboard when the command opens

Getting Started

Get an API key from Google AI Studio (Gemini, free tier available), the Anthropic Console, or the OpenAI Platform.
Run the Translate command and paste the key when prompted. With Provider left on Auto-Detect, the extension recognizes Gemini (AIza…), Anthropic (sk-ant-…), and OpenAI (sk-…) keys.
For an OpenAI-compatible server, open the extension preferences, set Provider to OpenAI-Compatible Server, and fill in API Server URL (for example https://openrouter.ai/api/v1 or http://localhost:11434/v1) and Translation Model. Local servers that need no key accept any value in API Key.
(Optional) In the preferences, set default languages, override the translation or speech model, choose the macOS system voice for pronunciation, or enable clipboard suggestions on open.

Provider	Default translation model	Pronunciation
Gemini	`gemini-3.5-flash-lite`	`gemini-3.1-flash-tts-preview`
Anthropic	`claude-haiku-5-5`	macOS system voice
OpenAI	`gpt-6-luna`	`gpt-4o-mini-tts`
OpenAI-compatible	none — set Translation Model	macOS system voice

OpenAI speech overrides tts-1 and tts-1-hd use the macOS system voice to preserve the selected pronunciation language. Pronunciation is unavailable when that language has no system voice.

Commands

Command	Description
Translate	Translate a word or short text
Translation History	Open your saved translations
View Flashcards	Review words you've saved

Supported Languages

English, Ukrainian, Polish, German, French, Spanish, Italian, Portuguese, Dutch, Czech, Swedish, Japanese, Korean, Chinese, Turkish, Russian, Belarusian

Keyboard Shortcuts

Shortcut	Action
`⌘ C`	Copy translation
`⌘ ⇧ L`	Pronounce source word
`⌘ ⇧ R`	Pronounce translation
`⌘ 1` → `⌘ 5`	Pick a word sense and save it to history
`⌘ ⇧ H`	Open History (from Translate)
`⌘ ⇧ T`	Toggle source and target languages
`⌘ D`	Delete entry (from History)
`⌘ ⇧ D`	Clear all history
`⌘ E`	Export history as JSON
`⌘ ⇧ A`	Export word history for Anki
`⌘ ⇧ Q`	Export word history for Quizlet

Development

Use the Node version pinned in .nvmrc — Node 22.22.2, which bundles npm 10.9.7 and matches the current Raycast extension toolchain. Then run the extension in Raycast dev mode:

nvm use        # Node 22.22.2 / npm 10.9.7
npm install
npm run dev

Regenerate package-lock.json only on npm 10.x. npm 11 records optional peer dependencies differently, producing a lockfile that passes locally but fails the store CI's npm ci.

License

MIT

Versions4

More AI providers, clearer word details, and Anki export - October 19, 2026

Translation now works with Gemini, Anthropic, OpenAI, or any server that speaks the OpenAI Chat Completions API, such as OpenRouter, Ollama, or LM Studio. With the new Provider preference left on Auto-Detect, the provider is recognized from your API key.
The Gemini API Key preference is now API Key, so Raycast asks for your key once after this update. Paste the same Gemini key to keep using Gemini.
Pronunciation uses OpenAI speech with an OpenAI key, Gemini speech with a Gemini key, and the macOS system voice for Anthropic and OpenAI-compatible servers. OpenAI's legacy tts-1 and tts-1-hd models also use the system voice to preserve the selected language. A new Pronunciation preference switches every provider to the macOS system voice. Starting any word or translation pronunciation, or leaving its view, stops speech already playing.
A word with more than five senses now shows the first five instead of failing the lookup.
Word detail panes now lead with the translation itself, set as the largest text in the pane, above a single dictionary line carrying the word you looked up, its pronunciation transcription, part of speech, key inflected forms, and a usage label such as formal or literary where one applies. Each element appears only when it exists for that word, so multi-word expressions and languages without a transcription convention simply show less.
Translations now request pronunciation transcriptions, inflected forms, and usage labels per sense, so a word that differs by part of speech — English record as a noun and as a verb — shows the transcription and forms belonging to the sense you picked.
Example sentences in the translate, history, and flashcard detail panes now show the source-language sentence first, with the target-language rendering beneath it. Empty stored examples are omitted without leaving formatting markers.
The looked-up word is now shown in bold inside the source-language example sentence, including inflected forms and mixed-script words such as Japanese Tシャツ in sentences without spaces. Han/kana lookups highlight only their exact text, even beside digits or Latin letters. Highlighting in spaced scripts requires a word boundary before the lookup, so unrelated fragments such as cat inside scatter stay plain.
Detail panes now list the pronunciation actions and their shortcuts beside the entry, and leave the row out for a language that neither your provider nor the macOS system voice can speak.
Pronunciation moved off ⌘O and ⌘⇧O onto ⌘⇧L for the source word and ⌘⇧R for the translation, in the translate, history, and flashcard views — the old chords are the ones Raycast reserves for Open and Open With.
Anki export now writes a two-column Front / Back file tagged as Anki's Basic note type, with the translation, its part of speech, and both example sentences formatted on the back of the card. Earlier exports used five separate columns, so a note type built from one of those will not line up — import into a fresh deck or remap the fields once.
Simplified translation and text-to-speech model preferences to accept any model ID your provider offers, without relying on hardcoded dropdown options. Leave the translation model empty to use Gemini 3.5 Flash-Lite, Claude Haiku 5.5, or GPT-6 Luna for your provider; leave the speech model empty for the provider's speech default.
Removed the separate reasoning preference so reasoning behavior follows the configured model.
Added an API Server URL preference so translation and pronunciation can be routed through a proxy or gateway, or to an OpenAI-compatible server, where it is required along with Translation Model. Enter a server root or a full API base such as https://openrouter.ai/api/v1. Your API key is sent to the host you configure, so plain http is accepted only for local addresses and credential-bearing URLs are rejected.
API requests reject redirects so your API key stays with the server you configured.
Reworked the translate screen while a translation is in progress: the word you are translating now appears as a row in the Translation section where its result will land, and your Recent translations stay on screen instead of the list clearing to an empty placeholder.
Word input now auto-translates after 0.7s instead of 1.5s.
Returning to a word already looked up in the same Translate session now restores its senses immediately, including after undoing a typo, without another API request. Cached results are kept separate for each language pair, provider, model, and API server.
Clipboard suggestions normalize extra spaces in short phrases before lookup and saving.
Failed exports and failed pronunciations now explain what went wrong instead of showing the raw system error and your file paths.

Configurable models and reasoning - June 8, 2026

Added a Reasoning Level preference for translations, defaulting to Minimum for the fastest Gemini responses.
Replaced the free-form translation model default with a curated Gemini model dropdown plus an advanced custom model override.
Replaced the free-form text-to-speech model default with a curated Gemini TTS dropdown plus an advanced custom model override.
Refreshed the extension icon.
Tightened the pronunciation prompt for more accurate text-to-speech output.

Switch language pairs, pronunciation, and configurable models - June 1, 2026

Added an in-command language pair selector: switch the source and target language straight from a dropdown in the Translate, Translation History, and View Flashcards commands, without opening preferences. The selection persists across commands.
Renamed the Source Language / Target Language preferences to Default Source Language / Default Target Language — they now apply until you pick a pair from the dropdown.
Added Translation Model and Text-to-Speech Model preferences so you can point to a newer Gemini model if a default is deprecated.
Reworked error handling into clearer, more consistent toasts, including deprecation-aware messages that tell you when a model is no longer available and how to switch.
Added a target-language purity rule so translations stay in the chosen language, reducing cases where words from another language leak into the output.
Reject non-existing/gibberish words instead of hallucinating translations.
Require example sentences to contain the original word, not a synonym.
Add post-validation that filters out senses with missing words in examples.
Handle phrasal verbs and idioms (e.g. "give up", "break the ice") as single vocabulary items.
Added word pronunciation via Gemini TTS (⌘O for source word, ⌘⇧O for translation).

Initial Release - April 8, 2026

Added Translate command to translate words and short texts between languages using Gemini.
Added Translation History command to browse saved translations.
Added Flashcards command for spaced-repetition review of saved words (SM-2 algorithm).
Added configurable source and target language support (17 languages).
Added color-coded part-of-speech chips: nouns (blue), verbs (red), adjectives (green), adverbs (magenta), prepositions (yellow), pronouns (purple), conjunctions (orange).
Added typo correction — misspelled words are auto-corrected before translation.
Added short text and sentence translation support with detail panel.
Added extension preferences for Gemini API Key, source/target language, and safe clipboard prefill behavior.
Added auto-translate with 1.5s debounce for word input, manual submit for text.
Added abort of in-flight requests on search text change.

Compatibility

macOS

VocaBuilder

VocaBuilder

Features

Getting Started

Commands

Supported Languages

Keyboard Shortcuts

Development

License

Commands3

Versions4

More AI providers, clearer word details, and Anki export - October 19, 2026

Configurable models and reasoning - June 8, 2026

Switch language pairs, pronunciation, and configurable models - June 1, 2026

Initial Release - April 8, 2026

Subscribe to our newsletter.

VocaBuilder

VocaBuilder

Features

Getting Started

Commands

Supported Languages

Keyboard Shortcuts

Development

License

Commands3

Versions4

More AI providers, clearer word details, and Anki export - October 19, 2026

Configurable models and reasoning - June 8, 2026

Switch language pairs, pronunciation, and configurable models - June 1, 2026

Initial Release - April 8, 2026