Free, open-source, fully local AI voice dictation for macOS. The offline alternative to Wispr Flow, Willow, and Aqua Voice.
Phemy is a completely offline macOS menu bar app that turns your voice into polished, LLM-optimized text. Press a hotkey, speak your thoughts, and Phemy transcribes with local Whisper, refines with a local LLM, and pastes the result directly into whatever app you're using — no cloud, no API keys, no subscription, no data leaves your machine.
Think of it as Wispr Flow or Willow, but free, open-source, and fully private. No usage limits. No word caps. No account required. Everything runs on your Mac.
Most AI dictation tools charge $8–15/month and send your voice to the cloud. Phemy does the same thing for free, forever, with zero network requests.
| Phemy | Wispr Flow | Willow | Aqua Voice | SuperWhisper | |
|---|---|---|---|---|---|
| Price | Free | $12/mo | $12/mo | $8/mo | $5/mo |
| Open source | Yes | No | No | No | No |
| Fully offline | Yes | No | No | No | Partial |
| Voice → LLM prompt | Yes | Yes | Yes | Yes | No |
| Auto-paste | Yes | Yes | Yes | Yes | Yes |
| No account required | Yes | No | No | No | No |
| Your data stays local | Yes | No | No | No | Partial |
Alt+Space Local Whisper Local LLM Auto-Paste
┌──────────┐ ┌──────────────┐ ┌──────────────┐ ┌───────────┐
│ Speak │ ──▶ │ Transcribe │ ──▶ │ Optimize │ ──▶ │ Paste │
│ freely │ │ on-device │ │ on-device │ │ anywhere │
└──────────┘ └──────────────┘ └──────────────┘ └───────────┘
- Press the hotkey (default:
Alt+Space) — a floating overlay appears - Speak naturally — watch the audio-reactive ring respond to your voice
- Release or press again — Phemy transcribes and optimizes your speech
- Press Enter — the polished text is pasted into the focused app
The entire pipeline runs on-device. Your voice data never leaves your Mac.
- Download
Phemy.dmgfrom the latest release - Open the DMG and drag
Phemy.appto/Applications - Launch Phemy from your Applications folder
- macOS may show "app from an unidentified developer" — right-click → Open to bypass
Requires macOS 14+ on Apple Silicon.
xcode-select --install
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
brew install cmake
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh # restart Terminal after
cargo install cbindgengit clone https://github.com/HyperNoodlez/phemy.git
cd phemy
./build-rust.sh # first build takes several minutes
swift run- Click the menu bar icon → Show Settings
- Transcription → download
base(142 MB) - LLM → download
Qwen2.5 1.5B(1 GB, fastest) orQwen3 4B(2.7 GB, best quality) - Press
Alt+Space, speak, press Enter to paste
Every component runs locally — Whisper for speech-to-text, a quantized Qwen LLM for text refinement. No internet connection required after model downloads. No telemetry. No analytics.
Trigger Phemy from any app with a system-wide keyboard shortcut. The transparent floating overlay shows real-time audio visualization without stealing focus from your current window.
- Toggle mode — press once to start, press again to stop
- Push-to-talk — hold to record, release to stop
- Customizable hotkey (any modifier + key combination)
| Mode | Description |
|---|---|
| Clean | Remove filler words, fix grammar, preserve intent |
| Technical | Precise technical terminology, clear requirements |
| Formal | Professional language, business-appropriate tone |
| Casual | Clean but conversational, friendly voice |
| Code | Structured coding task with language and requirements |
| Verbatim | Minimal cleanup, closest to original wording |
| Raw | No LLM processing — direct transcript output |
| Custom | Use your own system prompt |
- Cmd+V — standard paste (default)
- Cmd+Shift+V — plain text paste (strips formatting)
- Shift+Insert — alternative paste
- Type Out — simulates keystrokes character-by-character
Configurable delay (0–500ms) lets focus return to the target app before pasting.
Six color themes that tint the entire UI — buttons, badges, the recording ring, and mode cards:
Purple · Blue · Teal · Rose · Orange · Emerald
Every recording is saved locally with the raw transcript, optimized text, mode used, and duration. Browse and delete entries from the History tab.
Downloaded from ggerganov/whisper.cpp on first use.
| Model | Size | Speed | Notes |
|---|---|---|---|
| tiny | 75 MB | Fastest | Quick tests |
| base | 142 MB | Fast | Recommended |
| small | 466 MB | Medium | Better accuracy |
| medium | 1.5 GB | Slow | High accuracy |
| large-v3 | 3.1 GB | Slowest | Maximum accuracy |
Q4_K_M quantized GGUF models from the Qwen family, running via llama.cpp with Metal GPU acceleration on Apple Silicon.
| Model | Size | Notes |
|---|---|---|
| Qwen3 4B Instruct | 2.7 GB | Best quality |
| Qwen2.5 3B Instruct | 2.0 GB | Great balance of speed and quality |
| Qwen2.5 1.5B Instruct | 1.0 GB | Smallest and fastest |
~30–50 tokens/sec on Apple Silicon. Prompt optimization completes in under 2 seconds.
How does this compare to macOS built-in Dictation? Built-in Dictation gives you a raw transcript. Phemy goes further — it cleans up filler words, fixes grammar, reformats for context (code, email, technical docs), and auto-pastes the result. It's voice-to-polished-text, not just voice-to-text.
Does it work with ChatGPT, Cursor, Slack, etc.? Yes. Phemy pastes into whatever app has focus. It works everywhere you can type — ChatGPT, Cursor, VS Code, Slack, Gmail, Notion, Terminal, or any other app.
How accurate is the transcription?
Phemy uses OpenAI's Whisper models running locally. The base model is fast and accurate for everyday use. The large-v3 model matches state-of-the-art cloud transcription accuracy.
Does it work without internet? Yes, 100%. After you download the models (one-time), Phemy never makes a network request. Airplane mode, no Wi-Fi, air-gapped — it all works.
What languages are supported?
Whisper supports 99 languages. Set the language code in Transcription settings (e.g., en, es, fr, de, ja, zh).
MIT License. See LICENSE for details.
Built by Lab.gargé