How It Works
Three steps. Zero friction.
Pure speed. Pure privacy.
Press Hotkey
Hit Ctrl+` from any app. No window switching.
Ctrl + `Speak Naturally
Any language. Whisper translates to English automatically.
Watch It Happen
Words appear in Claude Code instantly. Under 1 second.
Speed Comparison
Average words per minute for code prompts
Typing 40 WPM vs Speaking 150 WPM. More context to AI. Less effort. Better results.
Features
Everything you need.
Nothing you don't.
100% Offline & Private
Whisper.cpp runs entirely on your machine. Your voice never leaves. No API keys, no servers, no logs. Ever.
Metal GPU Accelerated
Runs on your GPU via Apple Metal. Large model (2.9GB) transcribes in under 1s. Way faster than any CPU-based tool.
99 Languages → English
Think in Russian, Hebrew, Arabic, or any language. Get English output. Save 40–65% on Claude tokens per prompt.
Native Agent Integration
Built for Claude Code, OpenCode, Cursor, Windsurf, iTerm2. Text is injected directly — no copy-paste needed.
Screenshot Attachments
Drag & drop up to 18 screenshots. Show the AI your UI bug, error, or design — don't just describe it.
macOS + Linux
Native on Apple Silicon (M1–M5) and Linux (Ubuntu, Debian, Arch). NVIDIA CUDA and AMD ROCm supported on Linux.
3 Whisper Models
Tiny (500MB, <300ms) · Medium (1.5GB, ~700ms) · Large-v3 (2.9GB, <1s on GPU). Pick your speed/accuracy tradeoff.
Edit Before Send
Review and edit transcription before it's sent. Full history. Smart VAD auto-stops when you pause (1–30s timeout).
Language Support
Think in your language. Code in English.
Real test results from Whisper large-v3 on Apple M5 GPU. We show what works, what's partial, and what's not ready yet. No marketing fluff.
test_methodology: macOS TTS voices → afconvert 16kHz mono → whisper-cli -m ggml-large-v3.bin -t 1 -dev 0 on Apple M5 GPU. Token multipliers calculated via BPE cl100k analysis. Accuracy tested with developer-relevant phrases (TypeScript, array sorting, error handling). Real users with native accents may get different results. We update this table as we run more tests.
vs. The Alternatives
Why not Superwhisper?
| Feature | Voice2Agent← YOU ARE HERE | Superwhisper | Wispr Flow | macOS Dictation |
|---|---|---|---|---|
| Price | $5/mo | $10/mo | $12/mo | Free |
| 100% Offline | ||||
| Metal / GPU Accelerated | ||||
| Large Model Fast (<1s) | ~10s CPU | |||
| Claude Code — Native | NATIVE | |||
| Auto → English | ||||
| Token Cost Saving | ||||
| Linux Support | ||||
| Screenshot Attach | ||||
| No Subscription | Lifetime |
* Superwhisper Large model runs on CPU (~10s). Voice2Agent Large-v3 runs on Metal GPU (<1s).
Social Proof
Developers who switched to voice
Real feedback from the early-access community
Pricing
Simple pricing. No surprises.
Start free · 7-day trial · No credit card
Cancel anytime
- All Whisper models
- Unlimited transcriptions
- Screenshot attachments
- Metal GPU acceleration
- macOS + Linux
$3.25/mo · 2 months free
- Everything in Monthly
- Save 35% vs monthly
- Priority support
- All future updates
Pay once, own forever
- Everything in Annual
- Lifetime updates
- 2 devices included
- Priority support forever
Secure checkout via Stripe · 30-day money-back guarantee · Cancel anytime
FAQ