OPEN SOURCE VOICE PROTOCOL

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

Orchestrate autonomous agents, terminal runtimes, and desktop workflows hands-free. Powered by on-device Whisper with real-time spoken feedback, active listening and zero cloud dependencies.

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

Choose Your Character

Viv

🇺🇸 US English Expressive & Modern

Christopher

🇺🇸 US English Authoritative & Calm

Emily

🇮🇪 Irish English Gentle & Melodic

Aria

System Designer

🇺🇸 US English Crisp Executive

Sonia

Code Reviewer

🇬🇧 UK English Warm & Analytical

Stefan

🇺🇸 US English Fast & Energetic

Feedback Loop

ProActive Listening

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

VoiceFi Mobile Headset Companion

9:30

VoiceFi Mobile

STANDBY

Ready to Listen

TRANSCRIPT STREAM 00:00

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

HUD macOS

VoiceFi Ready (⇧⌘N)

🤖

Standing by • Dictate (⌃T) or speak to agent (⌃R)

Terminal

voicefi-daemon

VAD:

IDLE

[00:00.000] [BOOT] VoiceFi Core v0.8.4 initialized (darwin-arm64, Apple Metal ANE)

[00:00.012] [AUDIO] CoreAudio HAL default input: Built-in Microphone (48000Hz, 2ch)

[00:00.024] [VAD] Silero VAD v5 engine ready (energy_threshold=0.042, latency=12ms)

[00:00.038] [ASR] Faster-Whisper on-device Metal pipeline loaded (~80ms budget)

[00:00.052] [TTS] Neural Persona: Viv (Companion) • 0ms Apple Ava fallback standby

[00:00.068] [HUD] Native macOS capsule window mounted (480x58 @ top-right)

[00:00.082] [IPC] Unix domain socket bound: /tmp/voicefi.sock (0.00ms roundtrip)

[00:00.095] [HOOK] IDE Agent integrations active: Antigravity, Claude Code, Cursor

[00:00.110] [SERVER] HTTP & WebSocket companion bridge listening on 127.0.0.1:5141

[00:00.125] [SAFE] Acoustic Safe-Mode enabled (auto barge-in thresholding active)

[00:00.140] [READY] Autonomous feedback loop standby. Waiting for turn completion...

Local is Faster

From spoken thought to IDE code execution. Every bar is coloured by what kind of work the time is spent on — on-device compute, network transit, remote inference, or human friction — across local agent workflows, local push-to-talk tools, cloud speech APIs, and manual dictation.

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

On-device compute Network transit — audio leaves your machine Remote inference Human / UI friction

VoiceFi Local Agent Pipeline Metal CoreML + Direct IPC

120ms ⚡ Instant

Audio is transcribed in RAM via Apple Silicon Metal acceleration and dispatched directly to IDE subagents via local IPC. Every millisecond is on-device — zero cloud round-trips, zero queue delays.

Superwhisper / MacWhisper Local Whisper + Clipboard Paste

800ms 6.7× Slower

Also fully on-device and zero-egress — the honest local comparison. The cost is push-to-talk: focus the caret, hold the key, wait, then paste. No agent dispatch.

Cloud Speech Pipelines HTTP / Remote GPU

1,450ms 12× Slower

TLS handshake + WAN audio upload + remote ASR GPU queue + cloud LLM formatting + webhook roundtrip back to your machine. Most of the bar is red and amber: your audio, off your machine.

Manual Dictation & Context Switching Baseline, not a product

2,800ms+ 23× Slower

Switching window focus, holding push-to-talk, waiting for text, copy/pasting into the IDE prompt. Measures human friction rather than any single product.

Agentic Differentiators

One-Way Dictation vs. The VoiceFi (vifi) Feedback Loop

Traditional dictation tools only replace typing. VoiceFi connects the entire circular feedback loop with your autonomous AI agents.

Capability / Vector	VoiceFi (vifi)	Wispr Flow	Superwhisper / MacWhisper	ElevenLabs Cloud
Two-Way Feedback Loop Agent speaks turn soundbite → Auto-listens → Developer replies	✓ Continuous 2-Way Loop	✗ 1-Way Dictation only	✗ 1-Way Dictation only	✗ Disconnected from IDE
AI Agent Hook Integration Native stop & notification hooks for Antigravity, Claude Code, Cursor, Aider	✓ Native 1-Line Setup	✗ No Agent Hooks	✗ No Agent Hooks	✗ No IDE Agent Hooks
Hands-Free Auto-Listening (VAD) Mic arms automatically when agent finishes a turn (zero clicking)	✓ Automatic Turn-Based VAD	✗ Push-to-Talk only	✗ Push-to-Talk only	⚠️ Continuous Cloud Stream
Smart Turn Summarization Cleanses raw terminal logs into 1-sentence spoken briefings	✓ Intelligent TTS Briefing	✗ No TTS / Readout	✗ No TTS / Readout	✗ Generic Voice Output
Audio Privacy & Egress Local Metal/CoreML Whisper in RAM vs cloud audio streaming	🔒 100% On-Device (0 Egress)	⚠️ Cloud Audio Upload	🔒 0 Bytes Egress (Offline)	⚠️ Cloud Audio Streaming
Cost & Licensing Open source CLI freedom vs monthly recurring software subscriptions	💚 100% Free MIT CLI	$12 – $20 / month	$20 – $30 one-time / sub	$0.05 – $0.30 / min API

Looking for deeper technical & latency specs? Explore full stage-by-stage latency breakdowns, Unix socket IPC benchmarks, and runtime matrices.

Full Architecture & Benchmark Matrix →

Universal Compatibility

Agent & IDE Ecosystem

Integrates into your existing terminal workflows and editor environments.

Antigravity

Auto-Hooked

Claude Code

Supported

Cursor

Supported

VS Code

Supported

Ollama

Supported

Zsh / Bash

Native