🎙️ VoxPilot

Talk to your IDE coding assistant. 3 model families, 90+ languages, streaming transcription. 100% on-device.

Install from Open VSX View on GitHub

Ctrl+Alt+V

🎙️ "Create a REST API endpoint for user authentication using JWT"

📝 Transcribed locally — auto-capitalized, auto-punctuated

💻 Sent to Copilot Chat → code generated

Why VoxPilot?

🔒

100% On-Device

Your audio never leaves your machine. No API keys, no cloud calls, no telemetry. Privacy by design.

🧠

3 Model Families

Moonshine (fast), Whisper (90+ languages), Parakeet (real-time streaming). Pick the right tool for the job.

🤖

Works With Any Assistant

GitHub Copilot, Continue, Kiro, Cody — any VS Code chat participant. Just configure and talk.

🎙️

Voice Commands

Say "undo", "save", "comment", "function", "rename", "commit". Editor actions, code constructs, and git ops by voice.

🔇

Neural Noise Reduction

Auto-calibrating noise gate + RNNoise WASM denoiser. Works in noisy offices, cafes, and open floor plans.

📖

Auto-Vocabulary

Learns your codebase identifiers. "get user name" becomes getUserName. Scans open files automatically.

⚡

Smart Text Processing

Filler word removal, auto-capitalize, auto-punctuate, context-aware formatting. Transcripts arrive clean.

🎯

Flexible Output

Send to chat, editor, terminal, or clipboard. Walky-talky mode, wake word activation, idle auto-stop.

📊

History & Dashboard

Searchable transcript history, performance metrics, latency benchmarks. Track and improve your voice workflow.

🌍

Multi-Language & Cross-Platform

90+ languages with per-profile model suggestions. Linux, macOS, Windows. VS Code and Kiro.

Models

Browse, download, and switch models from the built-in Model Manager panel.

Model	Size	Strength
Moonshine Tiny	~27MB	Fastest, quick commands
Moonshine Base	~65MB	Fast, better accuracy
Whisper Tiny	~75MB	90+ languages, fast
Whisper Base	~150MB	90+ languages, balanced
Whisper Small	~500MB	90+ languages, high accuracy
Whisper Large v3 Turbo	~3GB	90+ languages, best accuracy
Parakeet TDT 0.6B	~150MB	Real-time streaming, live captions

How It Works

Microphone → Noise Gate → Adaptive VAD → ASR Model → Post-Processing → Delivery

🛡️ Privacy & Trust

🚫Zero telemetry — No usage tracking, no analytics, no phone-home

🔒Zero cloud — Audio processed entirely on-device via ONNX Runtime

🙅Zero data collection — No accounts, no sign-ups, no personal info

🔍Fully auditable — MIT-licensed, open source. Read every line

🎤Minimal permissions — Only microphone access. No network beyond model download

📜Privacy policy — Clear, human-readable policy with no legalese

—

Downloads

ASR models

90+

Languages

Cloud dependencies

MIT

Licensed

Get Started

Install, press Ctrl+Alt+V, and start talking. That's it.

Install from Open VSX View Source

Stay in the Loop

Weekly AI agent intelligence from the builder of VoxPilot.