v2.6.0 — words appear while you speak

Your voice,
with root access.

Hold a key. Speak. Whisper transcribes on your machine, an LLM strips the “um okay so basically”, and clean text lands at your cursor — in any app. Audio never leaves your device.

Download freeStar on GitHub
mac · windows · chrome  ·  $0 forever  ·  open source
⌥ hold-to-talk
0 cloud calls
~100ms cleanup
sudovoice — zshmic
$
scroll▼
● types into the apps you already use
VS CodeSlackChromeTerminalNotionGmailCursorWhatsAppFigmaObsidianDiscordLinearXcodeWord
0x
faster than realtime
# whisper on-device
0
cloud calls while you speak
# audio stays local
~0ms
LLM cleanup latency
# filler stripped live
$0
forever, for the core app
# AGPL-3.0 source
// 01 · what it does
$ echo $SPEECH | sudovoice

Speak at 150 WPM.
Read like you edited it.

you — speakingraw
“um okay so basically can you uh refactor this function to use async await instead of like promises”
→
sudovoice — writescleaned · 0.3s
Refactor this function to use async/await instead of promises.

press one key · start talking · clean, formatted text lands at your cursor

// 02 · try it in your browser
live — runs in your browser, no install

Try it right now.

Hit the mic and speak. Watch how fast voice beats typing.

voice-demo.txt

Click the microphone button below and start speaking...
Your words will appear here in real-time.

Uses your browser's built-in speech recognition (Chrome/Edge). No data sent anywhere.

// 03 · how it works
$ man sudovoice

One hotkey.
One pipeline.

pipeline — livestreaming
$mic|whisper|cleanup|cursor
# mic:
16kHz capture + voice activity detection
# whisper:
on-device transcription, 42x realtime
# cleanup:
LLM strips fillers, fixes grammar (~100ms)
# cursor:
clean text typed into the focused app
[1]

Hold the hotkey

Hold ⌥ Option from any app. The floating indicator lights up — you're live.

[2]

Speak naturally

Talk like a human, not a robot. Say “comma” or “scratch that” to edit as you go.

[3]

Release. It's typed.

Cleaned text appears at your cursor. done in 0.3s_

// 04 · features
$ sudovoice --help

Everything you need.
Nothing you don't.

--offline

Whisper runs on your machine

No internet required, no cloud processing, nothing to subpoena. Five model sizes from 75 MB to 3 GB.

network monitor▉ 0 B/s upstream
# all inference local — the graph above is your CPU, not your data leaving
--cleanup

The “um” never lands

An LLM rewrites your rambling into clean prose in ~100ms. Groq, Claude, OpenAI, Gemini + 3 more providers.

like you know the the server is is down again
→The server is down again.
--auto-type

Types at your cursor

VS Code, Slack, Chrome, Terminal — any app, any text field.

--voice-commands

Hands-free editing

“Scratch that” undoes. “Delete word”, “clear all” — no keyboard.

--punctuation

Smart punctuation

Say “comma” or “new paragraph” — the character, not the word.

--hindi

Hindi / Hinglish

Speak Hinglish, get clean English. Tuned for accents others ignore.

// 05 · time saved
$ time --spent-typing

Typing is your bottleneck.

You speak at 150 WPM and type at 40. Run the numbers on what the gap costs you.

⚠ the average professional loses 800+ hours/year to typing
Speed Comparison
⌨️ Typing speed40 WPM
🎙️ Speaking speed150 WPM
3.75x fasterwith your voice
3h
1h10h
5
1 day7 days
Based on industry averages: 40 WPM typing vs 150 WPM speaking (Natural Language Processing Institute)
Your time savings with voice
Weekly hours saved
11.0h
every single week
47h
per month
572h
per year
Value of time saved
$14,300
per year at $25/hr
That's $1,192/month you could reclaim
Start saving time now
// 06 · models
$ sudovoice --model

Pick your model. All free.

Downloads once, runs offline forever. Pick the size that fits your machine.

Tiny

Size: 75 MB
Speed: Fastest
Accuracy: Good
Best for: Quick notes, short phrases

Base

Recommended
Size: 140 MB
Speed: Fast
Accuracy: Better
Best for: Daily use

Small

Size: 460 MB
Speed: Medium
Accuracy: Great
Best for: Professional work

Large V3 Turbo

Size: 1.5 GB
Speed: Slower
Accuracy: Excellent
Best for: Maximum accuracy

Large V3

Size: 3 GB
Speed: Slowest
Accuracy: Best
Best for: Research, long-form
// 07 · vs. the others
$ diff sudovoice everything-else

Why pay for less?

Same job, no subscription, and the source is public.

comparison.tsv
fieldsudovoiceWispr FlowBridgeVoicemacOS Dictation
price
$0
no limits
$8/mo
~$96/yr
$50/mo
~$600/yr
free
built-in
privacyon-devicecloudon-devicecloud (Apple)
hindi/hinglish[x][ ][ ][ ]
open source[x][ ][ ][ ]
llm cleanup[x][x][ ][ ]
voice commands[x][x][ ][ ]
// 08 · pricing
$ cat tiers.conf | head

Start free.
Upgrade when it earns its keep.

--free
$0forever · no credit card
  • [x] 60 min/day on-device transcription
  • [x] auto-type into any app
  • [x] voice commands + smart punctuation
  • [x] AGPL source — self-host everything
Download free
--pro
$12/mo · $99/yr · $249 lifetime
  • [x] unlimited transcription
  • [x] all 5 Whisper models + 7 LLM providers
  • [x] Gemini Live streaming + batch files
  • [x] up to 5 devices
See full pricing

30-day money-back · lifetime capped at the first 100 buyers

// 09 · download
$ sudovoice --install

Install once. Own it forever.

Grant two permissions and you're dictating in 60 seconds. No account required.

macosv2.6.0 · 4 MB
os: macOS 14 Sonoma+
chip: Apple Silicon & Intel
ram: 4 GB min (8 GB for Large)
network: first model download only
Download .dmg
windowsv2.6.0 · ~81 MB
os: Windows 10 / 11 (x64)
cpu: x64; ARM via emulation
gpu: Vulkan → 6× speedup (optional)
network: first model download only
Download .exe
[ linux: coming soon ][ chrome extension: in review ]
⚠ warn: “Apple could not verify…”

Normal for apps not yet notarized by Apple. SudoVoice runs fully on-device — one-time approval:

  1. 1. Open the app → macOS blocks it → click Done
  2. 2. System Settings → Privacy & Security
  3. 3. “SudoVoice was blocked” → Open Anyway

One-time only. On macOS 13 and earlier: right-click the app → Open → Open.

⚠ warn: “Windows protected your PC”

SmartScreen flags installers without established reputation. The installer is exactly what the button serves, checksummed each release:

  1. 1. Click More info
  2. 2. Click Run anyway
  3. 3. Installer proceeds normally

If it hangs: right-click the .exe → Properties → tick Unblock → Apply → re-run.

// 10 · faq
$ man faq

Questions, answered.

?Does my audio ever leave my device?
No — by default Whisper runs entirely on your machine and the network stays silent. Cloud only enters if you explicitly enable optional extras: an LLM cleanup provider, Gemini Live streaming, or signed-in transcript sync. All three are off until you turn them on.
?Is it actually free?
Yes. The free tier (60 min/day, on-device, auto-type everywhere) is free forever — no credit card, no trial clock. Pro adds unlimited use and every model/provider. And the whole codebase is AGPL-3.0, so you can self-host everything for $0.
?Which platforms are supported?
macOS 14+ (Apple Silicon and Intel) and Windows 10/11 today. A Chrome extension is in review, and Linux is on the roadmap.
?How is this different from macOS dictation or Wispr Flow?
Three ways: it's offline by default (Apple's dictation and most rivals route audio through the cloud), an LLM strips your filler words before the text lands, and there's no subscription required — the core app is free and open source.
?Does it handle Hindi and Hinglish?
Yes — speak Hindi or code-switched Hinglish and get clean English text out. It's tuned for accents most dictation tools ignore.
?Can I read the code?
All of it. SudoVoice is AGPL-3.0 on GitHub — the Mac app, the Android app, and this website. Audit the audio path yourself; that's the point.
// 11 · feedback
$ echo "your thoughts" | sudovoice

Talk back to us.

Bug, idea, or a kind word — type it, or fittingly, speak it.

Goes directly to the developer. No spam, no tracking.

// 12 · support
$ sudovoice --support

Maintained. Actively.

report_bug()

Something broke?

Open an issue on GitHub with the log line — fixes ship fast.

open →
request_feature()

Missing something?

Tell us what you need. The roadmap is built from these.

open →
join_community()

Follow along

Tutorials, release notes, and builds-in-public on YouTube.

open →
iupdates announced on youtube — the app checks for new versions automatically.
// join early access

Get root access
to your voice.

$ sudovoice --installfree forever · no credit card · open source