Short answer: enable Auto-translation to English in AICHE settings, press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux), speak in whatever language your brain thinks in, and AICHE delivers a clean English prompt to Claude Code, Cursor, ChatGPT, or whatever has focus. 99 input languages.
The Problem
You think in German, but Claude Code reads English. You think in Mandarin, but Cursor's chat box expects English. You think in Russian, Japanese, Spanish, Hindi, or any of the other 96 languages AICHE supports, but the AI agent on the other end speaks English.
So you translate in your head while typing. Not just the words - the technical vocabulary, the sentence structure, the confidence that your phrasing is correct. That translation step is slow, draining, and it narrows what you express. Complex architectural ideas get simplified because formulating them in English takes too much effort. Constraints get dropped. Context gets cut.
Speaking in your native language removes the translation burden from your brain and hands it to the pipeline.
How It Works
- Open AICHE settings.
- Enable Auto-translation to English (sometimes labeled "Translate to English" depending on your platform).
- Press your hotkey anywhere - terminal, browser, IDE, chat window.
- Speak naturally in your language. Full sentences, technical context, whatever you'd say to a colleague.
- AICHE transcribes the audio in the source language, then translates the output to English. Message Ready formatting applies: filler removed, punctuation added, paragraphs placed.
- Clean English text inserts at the cursor.
The AI agent reads a well-formatted English prompt. Your brain never switched languages.
99 Languages
AICHE's voice input supports 99 languages across all platforms - macOS, Windows, Linux, iPhone, iPad, Apple Watch, Android, Chrome extension, Obsidian plugin, and REST API. Language detection is automatic, or you can set a preferred language in settings.
Some of the languages developers commonly use:
| Language | Works with voice input | Works with auto-translate |
|---|---|---|
| German | Yes | Yes |
| Mandarin Chinese | Yes | Yes |
| Spanish | Yes | Yes |
| Japanese | Yes | Yes |
| Russian | Yes | Yes |
| Hindi | Yes | Yes |
| Portuguese (Brazilian) | Yes | Yes |
| Korean | Yes | Yes |
| French | Yes | Yes |
| Ukrainian | Yes | Yes |
| Turkish | Yes | Yes |
| Arabic | Yes | Yes |
Without auto-translate, the output stays in the source language. With auto-translate, the output is English regardless of what you spoke.
Technical Terms Stay Technical
When you speak German prose mixed with English code terms ("erstelle eine React Komponente die useState verwendet"), the pipeline handles both layers:
- German prose gets translated to English.
- Code identifiers, library names, and CLI flags pass through the Software Development profile (Pro) and come out as
useState,kubectl,--dry-run- not as translated words. - Your custom vocabulary entries (repo names, service names, internal tools) apply to the English output. If you added
OrgAuthGatewayto your vocabulary, it appears asOrgAuthGatewayregardless of what language you spoke around it.
The practical effect: you can think about architecture in your native language, include code-specific terms in English as you naturally would, and the output reads like a native English prompt with correct technical vocabulary.
Use Cases
Prompting AI agents. The primary use case. Speak a detailed prompt in your language - full context, constraints, architecture - and the AI reads clean English. More detail per prompt, fewer correction loops.
Code review comments. Write PR feedback in your language. The translated output is professional English your teammates read. The review happens faster because you're not drafting English sentences in your head.
Documentation. Dictate README sections, architecture docs, and inline comments in your language. The English output goes into the codebase. Especially useful for teams where the code is English but the team's working language is not.
Slack, email, and chat. Reply to English-speaking colleagues by thinking in your language and speaking. The translation handles the rest. No tab-switching to a translation tool.
Platform Notes
- Desktop (macOS, Windows, Linux): the app UI is English-only. Voice input works in all 99 languages.
- Mobile (iPhone, iPad, Android): the app UI is localized into 28 languages. Voice input works in all 99.
- Apple Watch: voice input works in all 99 languages. The Watch records and uploads directly over its own connection.
Auto-translate is an account setting. Enable it once, it applies everywhere.
Tips
Don't code-switch mid-sentence unless it's natural. If you naturally say "erstelle eine React Komponente", keep doing that. If you normally say the whole sentence in one language, don't force English terms in. The pipeline handles both patterns, but it works best when you speak the way you'd speak to a colleague.
Custom vocabulary helps with names. Your internal service names and repo names probably don't translate well. Add them to the 50-entry vocabulary so they pass through unchanged regardless of language.
Review the first few outputs. Auto-translation quality is high for most languages, but spot-check your first few prompts to make sure the technical nuance is landing correctly in English.
Related
- Voice Commands for AI Coding - the broader AI prompting workflow
- Custom Vocabulary - 50 entries for your stack
- Software Development Profile - code-tuned recognition
- Multilingual Voice Input - full feature page
Result: complex architectural prompts you used to simplify for English now arrive with full detail. Your brain thinks in the language it's fastest in. The AI reads clean English.
Try it now: enable Auto-translation in AICHE settings, open Claude Code or ChatGPT, press your hotkey, and speak one prompt entirely in your native language. Check the English output.