OPEN SOURCE AMBIENT VOICE PROTOCOL

Give a voice to your agents,
and agency to your voice.

Never track terminal tabs or babysit split-panes again. VoiceFi is the local-first ambient voice protocol and macOS layer that announces agent turn completions aloud, chimes, and auto-listens for your next instruction hands-free.

Download for macOS Read Protocol Spec

$ curl -fsSL https://vifi.sh | sh

High-Fidelity Audio Bus (48.0 kHz)

Audition Neural Voice Personas

Experience distinct acoustic personas configured for autonomous planner, debugger, and code review agents.

ANTIGRAVITY LEAD

Christopher

Deep Neural · 48.0 kHz

vifi voice set planner Christopher

SYSTEM DESIGNER

Aria

Crisp Executive · 48.0 kHz

vifi voice set designer Aria

CODE REVIEWER

Sonia

Warm British · 48.0 kHz

vifi voice set reviewer Sonia

DEBUGGER COPILOT

Guy

Energetic American · 48.0 kHz

vifi voice set debugger Guy

Autonomous Feedback Loop

Ambient Voice Loop Simulator

Step through the zero-click autonomous cycle connecting agent runtimes directly to your voice.

voicegency-daemon • stdout

VAD:

IDLE — Click Start Loop

[00:00.000] VoiceFi daemon listening on port 8765...

[00:00.040] IPC socket connected. Autonomous turn-completion hook standby.

LIVE SPOKEN BUFFER 0 tokens

> Waiting for speech input... _

⚡ Quick spoken test phrases (click to simulate or speak): Push-to-Talk active

9:30

VoiceFi Companion

Mobile Remote • Pixel 11 / PWA

STANDBY

🎙️ IDLE

Ready to Listen

Zero-click ambient voice sync active

TRANSCRIPT STREAM 00:00

“Agent finished unit tests. Say 'Deploy to staging' to dispatch...”

Google Pixel 11 • Companion Mobile Bridge (PWA)

Click to start interactive loop

Brain Dump → Production Artifact

The Voice Memo Buffer

Pace around your room and ramble for 3 minutes. VoiceFi decomposes your thoughts into verified PR checklists.

RAW STREAM TRANSCRIPT (2:14) vifi memo record

“Hey so let’s refactor the auth session handler because right now if a token expires during a WebSocket reconnection it silently drops the audio chunk. We should store an ephemeral sliding refresh token in secure keychain on macOS, emit a re-auth event on the event bus, and retry the audio frame buffer without dropping the microphone stream...”

⚡ Synthesized into 3 Architectural Layers

vifi memo synth ~/.voicegency/memos/memo_01.wav

✓ Ready for PR

[AUTH-01] Add ephemeral sliding refresh token store in macOS Keychain. [WS-02] Intercept WebSocket 401 reconnection frames & emit AUTH_RENEW. [AUDIO-03] Buffer pending microphone audio chunks into a 5-second ring buffer during renewal. [TEST-04] Add unit tests for token expiration during active audio streaming.

[Mic Stream] ──► [5s Ring Buffer] ──► [WebSocket Bus] ──► [Agent Runtime]
                        │                    ▲
                  (Token Expiry)             │
                        ▼                    │
                [macOS Keychain] ──► [AUTH_RENEW]

Artifact: ~/.voicegency/memos/memo_auth.md Auto-Pushed to Antigravity

REAL-TIME LATENCY BENCHMARK

Why 100% On-Device Matters

Eliminate network round-trips, TLS handshakes, and cloud concurrency queues. Compare end-to-end latency to agent execution.

Architecture

0ms 500ms 1,000ms 1,500ms 2,000ms 2,800ms+

VoiceFi Local Whisper Metal CoreML

120ms ⚡ Instant

Audio is processed 100% in RAM via Apple Silicon Metal acceleration. Zero network round-trips, zero cloud queues.

Cloud Speech APIs HTTP / WebSocket

1,450ms 12× Slower

TLS handshake + HTTP upload + remote GPU queue concurrency + websocket text stream response.

Manual Push-to-Talk Human Latency

2,800ms+ 23× Slower

Window focus switching, mouse cursor positioning, holding push-to-talk key, waiting, copy/pasting.

Pillar 01 // Trust & Integrity

The Local-First Audio Manifesto

Voice data is sacred. VoiceFi is engineered to guarantee zero audio data ever leaves your computer.

🔒

100% On-Device Neural Whisper

Runs offline on Apple Silicon Metal acceleration. Speech is transcribed directly in memory on your Mac and discarded immediately after insertion.

🚫

Zero Audio Telemetry or Training

Your voice is never stored on cloud servers, sold to third parties, or used to train public AI models. Verified by open-source MIT code.

🛡️

Encrypted Local IPC Audio Bus

The internal vifi daemon uses secure local unix sockets and localhost loopback. Zero external open ports.

Universal Compatibility

Agent & IDE Ecosystem

Integrates automatically into your existing terminal workflows and editor environments.

Antigravity

Auto-Hooked

Claude Code

Supported

Cursor

Supported

VS Code

Supported

Ollama

Supported

Zsh / Bash

Native

Give a voice to your agents, and agency to your voice.

Audition Neural Voice Personas

Christopher

Aria

Sonia

Guy

Ambient Voice Loop Simulator

VoiceFi Companion

The Voice Memo Buffer

Why 100% On-Device Matters

The Local-First Audio Manifesto

100% On-Device Neural Whisper

Zero Audio Telemetry or Training

Encrypted Local IPC Audio Bus

Agent & IDE Ecosystem

Give a voice to your agents,
and agency to your voice.