MIT OPEN SOURCE · BYOK · APPLE SILICON MACOS

Speak. Ship.

Free, open-source, BYOK native macOS AI voice dictation.
121ms latency, self-learning personal lexicon, 100% local privacy.

Apple Notarized · Zero Gatekeeper Warnings macOS 14+ (M1–M4) 12 MB Native · $0 Forever
Dictating · Cursor 121 ms
Right ⌥ / Fn to speak
RAW SPEECH Filler & Self-Correction Cleanup
DIRECT TO CURSOR OpenAI ASR + GPT-4.1
No Window Switching · Types Directly Into Any Focused Cursor
Cursor VS Code Terminal / Warp Slack Linear Notion Obsidian Xcode Mail / Gmail Discord WeChat / Feishu Chrome / Arc
BUILT FOR BUILDERS & HEAVY WRITERS

Zero bloat. Pure engineering.

Speed, total model freedom, personal vocabulary, and local-first privacy—all dialed to 10.

01 / EXTREME LATENCY

4× faster than typing. Instant paste.

Pure native Swift 6 + C++ engineered to rival Typeless and Wispr Flow without the monthly tax.

TypeZero
180 WPM
Raw Dictation
Manual Fix
Keyboard
45 WPM
121msStreaming Finish
$0Subscription Fee
12 MBNative DMG Size
02 / BYOK · FREE FOREVER

Bring your own APIs. Mix freely.

Click nodes below to test pipelines. One OpenAI key powers both ASR and polishing.

Active Pipeline: OpenAI gpt-4o-mini-transcribe → OpenAI GPT-4.1 / GPT-5
03 / PERSONAL LEXICON · 0.5ms

Fix a word once. Remembered forever.

Sub-millisecond phonetic recall + silent auto-learning whenever you manually correct text.

streaming asr → StreamingASREngine 0.4ms
sherpa on x → sherpa-onnx Auto-Learned
s e o growth → SEO Growth Boundary Safe
04 / HARDWARE KEY MATRIX

Left & right modifiers. Zero misfires.

Hardware-level isolation of Left/Right ⌥ ⌘ ⌃ ⇧ and Fn, plus select-text voice rewrite.

05 / ZERO MIDDLEMAN · 100% LOCAL PRIVACY

Keys in macOS Keychain. Instant 886MB FP32 offline failover.

No relay servers, no accounts, no telemetry. Click to simulate an instant network drop:

☁️ Direct Cloud ASR OpenAI / Groq / Seed-ASR 2.0
⇄
🔒 Local C++ SenseVoice FP32 886MB Full-Precision · Air-Gapped
Full-bandwidth audio buffered in RAM · API Keys in macOS Keychain · Lexicon & history 100% local
CRAFTED IN MILLISECONDS · v0.4.8

Obsessive millisecond details that feel effortless

From the exact millisecond you press the hotkey to auto-muting background audio, every edge case is engineered away.

0ms Hot-Mic Pre-Roll

Wakes the hardware mic on KeyDown and prepends 180ms of pre-roll audio so your opening word is never clipped.

0.1s Trailing-Syllable Guard

Keeps capture open for 100ms after stopping so the last two syllables land cleanly with zero sluggishness.

Auto-Mute External Audio

Instantly mutes music, videos, and alerts the moment you speak, then restores volume right after capture.

Smart Direct / Proxy Routing

Mainland endpoints bypass VPN automatically while OpenAI, Gemini, and Claude keep native proxy routing.

Two-Pass Long-Dictation Accuracy

Enables bidirectional second-pass rescoring and live dialog context for crisp punctuation across long speeches.

Verify-to-Activate Engine Cards

Expanding a provider never switches silently—live credential verification must pass before lighting up green.

HEAD-TO-HEAD

Why stop paying monthly for voice input?

Feature & architecture comparison (responsive card layout on narrow screens)
Capability TypeZero (MIT) Typeless Wispr Flow macOS Dictation
Pricing ✓ Free & Open Source Paid Subscription Paid Subscription Free
BYOK APIs ✓ 3 ASR + 7 LLM Engines ✕ Closed Managed ✕ Closed Managed ✕ None
Offline Failover ✓ 121ms + 886MB FP32 Cloud Required Cloud Required No AI Polishing
Personal Lexicon ✓ 0.5ms Recall + Auto-Learn Basic Word List Manual Dictionary ✕ None
Privacy & Relay ✓ 0 Relay · Keychain · Local Vendor Cloud Relay Vendor Cloud Relay Apple On-Device/Cloud
QUICK ANSWERS

Everything in 30 seconds

Is TypeZero really free forever? What do API calls cost?
Yes—MIT open-source with zero subscription fees or paywalls. Run 100% free offline with the embedded 886MB SenseVoice FP32 + local Ollama, or plug in your own OpenAI / Groq / DeepSeek / Gemini API key at raw provider rates (typically a few cents per day for thousands of words).
Can I start with just an OpenAI API key?
Absolutely. v0.4.5 ships with native OpenAI (gpt-4o-mini-transcribe) and Groq (whisper-large-v3-turbo) speech cards. Enter your OpenAI key once and TypeZero automatically reuses it for both voice transcription and LLM formatting.
Do my transcripts or API keys ever touch a developer server?
Never. There is no TypeZero backend relay, no user account, and zero telemetry. API keys are encrypted in macOS Keychain, requests go directly to your chosen provider, and the full codebase is open for audit.
Which Macs are supported? Will Gatekeeper block it?
Supports Apple Silicon (M1 / M2 / M3 / M4) running macOS 14 Sonoma or later. Every DMG release is signed with Apple Developer ID and Apple notarized & stapled—drag into Applications and launch with zero warnings.

Recent Updates

All Releases →
v0.4.8LATEST
  • Smarter paragraph flow
  • Auto-fold mid-speech edits
  • Chat, mail & IDE tone adapt
v0.4.7
  • Auto-mute audio on dictate
  • Instant volume restore
  • Direct China & Global mirrors
v0.4.6
  • 0ms hot-mic, no clipped start
  • 0.1s tail-syllable capture
  • VPN-bypass for domestic LLMs

Give typing back to your voice.

TypeZero v0.4.8 · Apple Notarized · 12 MB Native DMG · MIT Free Forever

brew install --cask https://raw.githubusercontent.com/nasadunadu/TypeZero/main/Casks/typezero.rb