Voice to Mac runs Whisper locally on Apple Silicon — instant, private, and engineered to work inside the editors you actually live in. No cloud, no subscription, no compromise.
Raw transcription happens on your Mac, always. AI cleanup is an optional second stage — skip it, bring your own keys, or use our managed service.
Five model sizes ship in-app, compiled with Metal acceleration. Tiny (39 MB) for quick notes, Large-v3 (1.5 GB) for near-cloud accuracy. No network, ever.
whisper.cpp · MetalStrips filler ("um," "you know"), fixes punctuation, and kills 16 known Whisper hallucinations like "Thanks for watching!" Still 100% local.
deterministic · offlineRoute transcripts through Claude, GPT-4o, Groq, or local Ollama for tone and formatting. Opt-in, one provider at a time, keys in your Keychain.
BYO keys or subscribePer-app injection strategies, a load-bearing paste delay for Electron, and a ~700-character bias prompt of your own identifiers. Speak "set loading to true" and it comes out setLoading(true).
Each of these exists because a real user hit a real edge case. The details are the product.
Transcription happens 100% on your Mac — no network calls, ever. The only automatic requests are a daily update check and opt-in crash reports; both can be disabled in Settings, so Voice to Mac can run fully offline. If you add an AI provider, your API keys live in the macOS Keychain, never in plain-text config.
One-time payment. Own it forever, all future updates included.
Try it free for 14 days. No card required.
No. Transcription happens on your Mac via whisper.cpp compiled with Metal. Even the optional AI cleanup sends text, never audio — and only if you explicitly enable it.
No. Voice to Mac is a one-time $19.99 payment — you own the license forever, all future updates included. You bring your own API keys for cloud AI cleanup (Anthropic, OpenAI, Google, Groq, xAI, Azure, custom) and pay those providers directly for usage, or run it 100% locally with Ollama or LM Studio.
Yes — 14 days, fully featured, no card required. Download, install, and dictate as much as you want. When the trial ends, a one-time $19.99 license keeps everything running forever.
macOS 13.0 or later on Apple Silicon (M1 or newer). Whisper needs Metal for real-time performance; we don't support Intel Macs.
Yes — this is the whole point. We maintain per-app injection strategies for 17 known apps, including Cursor, VS Code, Windsurf, Xcode, Linear, Notion, and Slack. A code-aware prompt feeds your own identifiers into Whisper so it recognizes domain-specific terms.
Whisper Large-v3 is the same model behind most commercial transcription APIs. On an M-series Mac it runs fast enough to feel instant, and our code-aware prompt builder materially improves accuracy on identifiers and technical vocabulary.
macOS 13+, Apple Silicon. 5 MB app, no signup required.
