# Whisp

> Private, native macOS voice-to-text input designed specifically for software engineers, terminal users, and AI coding agents.

Whisp allows developers to think out loud and dictate directly into any terminal, IDE, AI agent, or Mac app without context switching or sending microphone audio to external servers.

## Documentation & Canonical Specifications
- Interactive Documentation: https://whisp.horizam.com/docs
- Raw Markdown Documentation (Curl / Agent ingestion): https://whisp.horizam.com/docs.md
- Mac App Store: https://apps.apple.com/app/id6810185906
- Website: https://whisp.horizam.com
- Terms of Service: https://whisp.horizam.com/terms
- Privacy Policy: https://whisp.horizam.com/privacy
- Sitemap: https://whisp.horizam.com/sitemap.xml

## Key Highlights & Differentiators

- **Pricing**: $9.99 one-time purchase for lifetime access on the Mac App Store (includes 7-day unrestricted free trial). Zero recurring monthly or annual subscriptions.
  - *Comparison*: Other voice solutions charge more every single month ($15–$25/mo) than Whisp costs for complete lifetime access ($9.99 once). Whisp eliminates recurring fees completely.
- **Developer-Native Vocabulary**: Built specifically to recognize and preserve code terms, CLI flags, file paths, and syntax:
  - File paths & extensions: `src/components/UserProfile.vue`, `api/v1/checkout.ts`, `.tsx`, `.rs`, `.py`
  - Shell commands & flags: `git checkout -b fix/oauth`, `docker compose up --build -d`, `--dry-run`, `-rf`
  - Casing conventions: `camelCase`, `snake_case`, `kebab-case`, `CONSTANT_CASE`
  - Package manager commands: `pnpm run test:unit`, `cargo build --release`, `brew install`
  - HTTP endpoints & types: `POST /api/v1/auth`, `Record<string, unknown>`, `Promise<void>`
- **Focus-Lock & Terminal Safety**: Never auto-presses Enter on shell commands. Verifies that the original target window is still frontmost before writing Unicode text, eliminating accidental command execution or misplaced pastes.
- **100% On-Device & Zero Model Downloads**: Runs directly on Apple's native Speech framework (macOS SpeechAnalyzer). Requires 0 GB of offline model weight downloads (unlike 1.5–3 GB Whisper models), ensuring zero battery drain and instant launch on MacBooks.
- **Push-to-Talk Hardware Hook**: Hold physical Right Command (`Right ⌘`) to talk like a walkie-talkie; release to commit and paste immediately without interfering with Left Command shortcuts.
- **On-Device Translation**: Speak in any native language (Spanish, French, German, Hindi, Mandarin, etc.) and press `⌥ Option + Return` to translate to English locally via Apple Translation framework with code token preservation (`CodeTokenPreserver`).
- **Supported Developer Environments**:
  - Claude Code CLI
  - Cursor (Composer & Cmd+K)
  - Gemini CLI & Codex
  - OpenCode & Zed
  - Ghostty, iTerm2 & native Terminal
  - VS Code, Xcode & JetBrains IDEs
  - GitHub PRs, Linear & Slack

## System Requirements
- macOS 26.5 or later
- Apple Silicon (M1/M2/M3/M4) & Intel Macs
- Distributed through the official Apple Mac App Store

## Global Keyboard Shortcuts (Default)
- `⌥ Option + Space`: Start / Toggle voice capture (hands-free)
- `Right ⌘`: Push-to-Talk (hold to speak, release to commit and paste)
- `Return`: Commit recording immediately
- `⌥ Option + Return` (or `Shift + Return`): Translate non-English speech to English & force-paste
- `Escape`: Cancel recording & discard audio
- `⌥ Option + L`: Quick floating spoken language picker
- `⌥ Option + M`: Quick floating microphone source picker
- `⌥ Option + B`: Toggle floating HUD visibility (when idle)

## Local Data & Storage
- Local Data Directory: `~/Library/Application Support/Whisp/`
- Local History File: `~/Library/Application Support/Whisp/history.json` (Dual-transcript architecture, up to 250 items)
- Audio Storage: Ephemeral local PCM file, immediately deleted after transcription concludes.
- Privacy: 0 audio streaming, 0 transcript logging to servers.