100% Local, 100% Private
All inference runs on Apple's Neural Engine. No cloud calls, no accounts, no telemetry, no API keys. Your audio, transcripts, and vocabulary never leave your machine. Works offline. See our data retention policy.
Push-to-talk dictation built for developers using AI coding agents. Tuned Voice runs entirely on your Mac's Neural Engine and learns your technical vocabulary from your conversations with Claude Code, Codex, Cursor, Windsurf, Cline, and Aider. Your audio never leaves the device.
You talk to coding agents all day. Your dictation tool should understand what you're saying.
All inference runs on Apple's Neural Engine. No cloud calls, no accounts, no telemetry, no API keys. Your audio, transcripts, and vocabulary never leave your machine. Works offline. See our data retention policy.
Optimized for Apple Silicon's Neural Engine. Text appears as you speak, not after you stop. Press your hotkey, talk, release. Text appears wherever your cursor is — any terminal, any IDE, any app.
Tuned Voice reads your local AI coding conversations to learn the technical terms you actually use — project names, function names, libraries, frameworks. The more you use it with your coding agent, the better it gets. This approach is patented.
$20 one-time purchase. No subscription traps. No data harvesting. No investor pressure to monetize your voice. Built by one developer who uses this every day to build software with AI agents.
No subscription. No annual renewal. Pay once and own it.
Lifetime license, all future updates included
Struggle with dictation due to an accent or speech difference? We offer a custom voice model training service — a speech model fine-tuned specifically to how you speak, for dramatically better accuracy. Get in touch to learn more.
The only dictation tool that combines on-device privacy with vocabulary learning from your coding conversations.
| Feature | Tuned Voice $20 lifetime |
Superwhisper $249 lifetime |
Wispr Flow $15/mo |
|---|---|---|---|
| Works offline | Yes | Yes | No — requires internet |
| Audio sent to a server | Never | Never | Yes — every dictation |
| Requires API keys | No | Optional | Cloud account required |
| Learns coding vocabulary | Yes — from AI conversations | No | No |
| AI coding agent integration | Claude Code, Codex, Cursor, Windsurf, Cline, Aider | No | Partial |
| 100% on-device inference | Apple Neural Engine | Yes | Cloud-only |
| Audio saved by default | No — discarded after transcription | Saves recordings | Cloud processing |
| One-time purchase | $20 | $249 | $180/yr |
Tuned Voice uses a patented approach to learn your vocabulary from the AI coding conversations already on your Mac — automatically.
Works immediately with a high-quality on-device AI model. Push-to-talk dictation in any app — no setup, no account, no cloud, no API keys. Press your hotkey, speak, release. Text appears wherever your cursor is.
When you use Claude Code, Codex, Cursor, Windsurf, Cline, or Aider, Tuned Voice reads your local conversation history to identify the technical terms you actually use — project names, function names, libraries, frameworks, API endpoints. It builds a custom dictionary of terms and corrections automatically. Everything stays on your Mac.
During live dictation, Tuned Voice tracks confidence on every single word. Terms the model is unsure about get flagged and boosted — so your technical vocabulary gets recognized correctly, and accuracy improves with every sentence you speak.
Your custom dictionary and confidence weights work together so domain-specific terms are recognized correctly the first time. The more you use Tuned Voice with your coding agent, the better it gets. This conversation-learning approach is patented.
Tuned Voice is actively developed. We ship updates based on your feedback and want to hear what's working and what isn't. Tell us what you think.
You already talk to Claude Code, Codex, Cursor, and Windsurf all day. Tuned Voice makes sure every word is right.
Dictate prompts to Claude Code, Codex, Cursor, Windsurf, Cline, or Aider. Variable names, API calls, framework names, project-specific jargon — all recognized correctly because Tuned Voice learns from your conversations.
Works in any terminal — iTerm, Warp, Terminal.app, VS Code integrated terminal. Press your hotkey, speak, release. Text appears at the cursor. No mouse, no switching apps.
Native voice features in coding agents don't work over SSH. Tuned Voice does — dictate locally, text appears in your remote session. Your mic, your Mac, your model.
No audio leaves your machine. No cloud processing. No API keys. No telemetry. Tuned Voice has no network code in its inference path — your voice data physically cannot be exfiltrated.
Tuned Voice works in any app, any terminal, any IDE. It reads the conversation history already on your Mac to learn your vocabulary. Nothing is ever sent anywhere.
Reads your Claude Code conversation history locally. Technical terms, package names, and code patterns become part of your vocabulary automatically.
Learns from your Codex sessions and code generation outputs. The vocabulary from your autonomous coding workflows feeds directly into your dictionary.
Scans Cursor workspace context and chat logs on-device. Framework APIs, function names, and project-specific jargon get recognized on first try.
Works with any coding agent in any terminal or IDE. Just press your hotkey, speak, and the text appears wherever your cursor is. Tuned Voice learns from whatever tools you use.
Tuned Voice isn't just a dictation tool — it's how we build software.
After my kid was born, I couldn't sit at a keyboard all day anymore. I started using voice-to-text with Claude Code and never looked back.
Every feature in Tuned Voice — the dictation engine, the vocabulary learning pipeline, even this website — was built almost entirely by voice. I speak what I want, Claude Code writes it, and together we ship real software faster than I ever could typing.
That's what made me realize: voice-to-text isn't just convenient, it's a superpower. But only if the recognition is actually accurate for the words you use every day. That's why I built Tuned Voice — dictation that learns from your coding conversations.
brew install --cask shanemmattner/tunedvoice/tunedvoice
macOS 14+ · Apple Silicon (M1 or later) · 7-day free trial · $20 lifetime