ROARWindows
StableWindows release metadata is loading.
Windows Stable unavailable View ReleasesDownload the verified installer, then run it.
100% local · Verified releases below
Speak. It types.
Nothing leaves your machine.
ROAR is private dictation for Windows — built for engineers shipping code, students drowning in notes, and anyone whose day is mostly typing. Hold a hotkey, talk, release. Clean text lands in whatever app you're already in.
Free forever for everyday dictation. No account. Windows 10/11.
The speech model downloads once on first run; then dictation stays offline.
Hold Ctrl + Win and talk — or tap the fob and go hands-free
Runs 100% offline
No cloud transcription
No account
No telemetry
Verified releases
Download ROAR
Choose your platform. We never detect or preselect it for you.
Linux
PreviewLinux Preview availability is checked from verified releases.
Linux Preview unavailable View ReleasesRead the Linux Preview instructions before installing.
Who it's for
Built for people who type for a living
Or study for one. Same tool, three very different days.
For engineers
A voice layer that respects your editor.
- Code Mode — verbatim text, spoken symbols
- Per-app profiles: terse in the terminal, polished in email
- Voice rules that run hotkeys, scripts, even OSC cues
For students
Lectures, essays, and every language you speak.
- Meeting capture turns online lectures into searchable notes
- Translate mode: think in your language, type English
- Draft essays at speaking speed — filler words removed
For work
Professional prose, without the compliance headache.
- Clean, formal output for email and docs — automatically
- No cloud processor to run past security review — your words never leave the building
- One payment per seat — no subscription
Three superpowers
Dictate. Automate. Route.
Dictation is the front door. Flow is what's behind it.
01 / DICTATE
Speech in, finished writing out
Not a raw transcript — ROAR strips the "um"s, false starts and filler as you go, fixes the words it used to mishear because you taught it, and matches its tone to the app you're in. GPU-accelerated when you have one; ~half-second latency on a modern machine.
02 / AUTOMATE
Say it, and it happens
Flow rules turn phrases into actions: open an app, press a hotkey chord, insert a snippet, speak a reply, call a webhook, cue show-control gear over OSC. Rules only fire at the start of a dictation — never mid-sentence — and anything that runs code needs your explicit trust, per rule.
03 / ROUTE
One dictation, everywhere it belongs
Typing into the focused app is just the default. Say "roar route notes on" and every dictation also lands in a timestamped notes file. Add the clipboard. Have it read back aloud. Meeting capture even transcribes what your PC is playing — calls and lectures — into a live, searchable local transcript.
Everything else
Dictation that stays yours
Speak it, type it, route it, automate it — everything runs on your machine: the speech model, your history, your words.
Push-to-talk, anywhere
Hold your hotkey in any app — editor, browser, chat — speak, and release. The text lands in the focused field.
Fast, on your hardware
NVIDIA CUDA and AMD/Intel GPU acceleration when you have it, tuned CPU when you don't.
Reads like writing
Removes "um", stutters and false starts as you go, so what lands is clean prose — not a raw transcript.
"Scratch that"
Say it and the last thing you dictated is removed — no reaching for the mouse.
Speaks your language
Multilingual dictation built in — pick a language, the model fetches once, and from then on it's offline for good.
Your history, your rules
Keep a local history or none at all. Delete it whenever. Those controls are free, forever, in every edition.
Point & Speak
Highlight text anywhere — browser, PDF, editor — and Ctrl + Right-click to hear it. One gesture, no menu. The voice is generated on your own machine, not a cloud API, and it's GPU-accelerated when you have one.
Teach it how you talk
Got a word it always mishears? Fix it once and ROAR remembers — it corrects that word from then on, and listens for the right one next time.
Dictate in any language, type English
Flip on Translate mode and speak French, Spanish, Mandarin — ROAR types English. Translation happens in the same local model; nothing goes online.
Why local matters
The privacy promise
This isn't a marketing line — it's how the app is built.
- Your voice never leaves your machine. Transcription runs locally. There is no cloud path to disable — there isn't one.
- No telemetry, ever — and it never phones home. The only network traffic is what you explicitly trigger: the one-time model download, an update check you click, or a webhook/OSC rule you created yourself.
- Your licence is checked offline. No account, no sign-in, no licence server phoning home.
- Privacy is never paid. History and audio deletion, retention controls and offline use are free in every edition — permanently.
Pricing
Buy once. Keep it.
One-time purchase. No subscription, no account, no cloud. Your licence is verified on your own device.
ROAR Core
Fast, private Windows dictation that runs on your machine.
Free
- Local push-to-talk dictation
- Hands-free dictation
- Streaming preview
- Multilingual transcription
- Translate to English
- Local history
- Privacy and deletion controls
- Basic vocabulary
- No account required
Best for most people
ROAR Pro
Smarter formatting and workflow tools for everyday writing.
$19 once
- Everything in Core
- Smart formatting modes
- Advanced cleanup
- Snippet packs and variables
- Vocabulary suggestions
- Advanced history filters
- Settings import and export
- Flow voice rules + multi-output routing
- Meeting capture (system audio)
- All v1.x updates included
Best for developers
ROAR Developer
A local, code-aware voice layer for technical work.
$29 once
- Everything in Pro
- Code mode
- Spoken programming symbols
- App profiles
- Per-app formatting and language
- Project vocabulary
- Developer snippet packs and file tagging
- Scripted Flow actions — scripts, webhooks, OSC
- All v1.x updates included
ROAR Supporter
Everything in Developer, with extra support for ROAR's continued development.
$49 once
- Everything in Developer
- Same Developer feature access
- Supports independent development
- All v1.x updates included
- No exclusive features — it's the same product as Developer. Buy this only if you want to fund the work.
One payment. No subscription.
No account required · No cloud transcription · No telemetry ·
Offline license verification · Your purchased version keeps working ·
Includes all v1.x updates · Optional paid major upgrades may happen later.
Answers
Questions
Is Core a trial?
No. Core is free dictation you can use forever — it doesn't expire, nag, or lock up. The paid editions add workflow power on top; they don't unlock the basics.
Do I need an account or an internet connection?
No account, ever. The network is needed exactly once — to download the speech model on first run. After that, ROAR transcribes and verifies your licence entirely on your own device; dictation itself never touches the internet.
Is this a subscription?
No. It's a one-time purchase. All v1.x updates are included. A future major version may be an optional paid upgrade, and the version you bought will keep working.
What happens to my dictation history?
It stays on your machine in a local database — or nowhere at all, if you turn it off. You can delete it, and any retained audio, at any time. Those controls are free in every edition.
What's the difference between Developer and Supporter?
Functionally, nothing. Supporter has exactly the same capabilities as Developer. It exists for people who want to fund independent development — there are no exclusive Supporter features, and we won't invent any.
How do I activate a licence?
Your licence arrives as a small file. Open ROAR's Settings, go to License, choose the file, and you're done — it's verified on your own device, with no account and no internet connection. Licences are currently issued by hand after each purchase, so allow a little time; ROAR keeps working in the meantime.
Which GPUs are supported?
NVIDIA via CUDA out of the box. AMD and Intel GPUs are supported through an optional Vulkan engine you can enable in Settings. No GPU? A tuned CPU path is the default.