Get VoiceFi for Mac
Run the open-source CLI from GitHub, or install the native VoiceFi.dmg macOS app with 100% offline neural voice, Screenpipe vision, Hey Pal Teach Mode, and hardware-matched local LLMs.
Open Source
Terminal CLI and Model Context Protocol (MCP) server for developers who want to wire voice hooks from source.
vifi listen / vifi speak)say & Edge TTSDeveloper Pro
Everything in the VoiceFi.dmg installer: native macOS HUD, offline Kokoro-82M neural voice, Hey Pal Teach Mode, and hardware-matched Local Compute models.
tev1:0.8b + MLX Coder)Annual Special
1-Year Pro Pass with the complete VoiceFi.dmg Local Compute suite, 20+ voices, and priority model updates.
Qwen2.5-Coder-14B)Free GitHub CLI vs. VoiceFi.dmg Installer Bonus
The GitHub repository gives you the lightweight CLI & MCP bridge. Installing VoiceFi.dmg turns your Mac into a 100% Local Compute Voice & Workflow OS — bundling offline neural speech and screen perception inside a ~200 MB drag-and-drop app, then automatically downloading the exact local LLMs your Mac's Unified Memory can run in the background.
| Capability / Component | Free GitHub CLI pip / git clone |
VoiceFi.dmg Installer 1 Unified Access Level |
|---|---|---|
|
IDE Hooks & MCP Server
Antigravity, Claude Code, Cursor & Windsurf turn-end voice hooks
|
✓ Included | ✓ Auto-Linked on Launch |
|
macOS Menu Bar, Dynamic Island HUD & Quick Bar
Hardware-locked notch waveform, live transcript pill &
Ctrl+Space bar |
— | ✓ Bundled in .dmg |
|
Kokoro-82M FP16 Offline Neural Voice (
Viv)196 MB studio neural TTS embedded inside
VoiceFi.app — works 100% offline on first launch |
macOS say only |
✓ Bundled in .dmg (196 MB) |
|
Smart-Turn v3.2 Prosody + Native Apple Speech Stream
8.3 MB ONNX vocal prosody end-of-turn detector + 65 KB native Swift
SFSpeechRecognizer binary |
Manual setup | ✓ Bundled in .dmg |
|
Zero-Egress Screenpipe Daemon (
:3030) + Foveal AX DOMBuilt-in Apple Vision OCR +
<5ms AXUIElement accessibility inspection with zero external dependencies |
— | ✓ Bundled & Auto-Started |
|
Hey Pal Teach Mode & Tier-1 API Compiler
Granola-style red target highlighter, browser XHR/HAR capture & 3-Stage AST SecGate Python script compiler
|
— | ✓ Exclusive to .dmg |
|
141ms Reflex & HAR Traffic Cop (
tev1:0.8b)Extracts spoken voice parameters and filters noisy browser HAR captures into clean REST/GraphQL endpoints
|
— | ✓ Auto-Downloaded in Background (811 MB) |
|
4.97s Self-Healing Workflow Coder (
voicefi-coder-mlx)4-bit Direct-on-Metal MLX model that repairs broken UI selectors and applies live spoken workflow edits
|
— | ✓ Auto-Downloaded in Background (839 MB) |
|
2.15s PbE Regex & Workflow Compiler (
Qwen2.5-Coder-14B-4bit)Induces regex rules from demonstration examples & synthesizes multi-step Python automations (100% LocalBench)
|
— | ✓ Auto-Downloaded on 16GB+ Macs |
Bundled Inside VoiceFi.app
Kokoro-82M neural TTS, Smart-Turn v3.2 prosody, Silero VAD v6, Apple Speech stream, and the Screenpipe :3030 daemon work the second you drag VoiceFi.app to Applications.
Core Local Reflex & Self-Heal
On first launch, VoiceFi automatically downloads tev1:0.8b (141ms voice parameter extraction & HAR API triage) and voicefi-coder-mlx (4.97s workflow self-healing) in the background.
Pro Compiler & Studio Brain
Same single DMG access level — if your Mac has 16GB+ Unified Memory, VoiceFi also provisions Qwen2.5-Coder-14B-4bit on Metal for 2.15s neural regex induction & full Python synthesis.
Apple Silicon Native
Universal binary compiled natively for Apple Silicon (M1/M2/M3/M4) and Intel Macs on macOS 13+.
Menu Bar Companion
Live microphone audio level meter, voice persona picker, and instant hotkey toggle straight from your macOS menu bar.
100% Local Bus
Direct UNIX domain socket & localhost IPC. Your audio and code context never leave your machine.