Releases: xhan145/roar
Release list
ROAR v0.35.2 — installer honesty, continued
Installer honesty, continued. An adversarial review of the v0.35.1 packaging fixes found four real defects — two of them introduced by v0.35.1 itself. The app is unchanged; like v0.35.1, this release is entirely about installing and building.
Fixed
An install that needs a restart no longer claims it failed. Windows returns a distinct "installed, but restart to finish" result when it had to replace files that were in use. v0.35.1 reported that as FAILED — nothing was changed, which was untrue, and never mentioned restarting — leaving you free to launch a half-replaced app. It now confirms the install succeeded and tells you to restart before launching ROAR.
The build could delete the installer package it had just made. The v0.35.1 rule kept the two newest packages, which quietly discarded a freshly built one whenever two newer versions were already present — building an older version to reproduce a bug, for instance — while still printing "built" and reporting success. The version being built is now always kept.
The version stamp is written next to the build recipe rather than wherever the build was started from, so it can neither go missing nor be stamped from a stale copy.
Upgrading
Upgrading in place from v0.35.1 works normally. If you are on v0.34.0 or earlier and an update appears to do nothing, uninstall ROAR from Settings → Apps first, then run this installer. Your data is safe either way — settings, licence, history, snippets and vocabulary live outside the program folder and are never touched by install or uninstall.
Your existing licence and any trial in progress carry over untouched.
Full changelog: https://github.com/xhan145/roar/blob/main/CHANGELOG.md
ROAR v0.35.1 — upgrades that actually upgrade
Upgrades that actually upgrade. This is a packaging-only release — the app itself is unchanged from v0.35.0. If v0.35.0 installed cleanly for you, there is nothing new here. If updating appeared to do nothing, this is the release that fixes it.
Fixed
Installing a new version could silently leave the old one in place. An in-place upgrade could finish, report success, and change nothing — same version, same files, no error. ROAR.exe now carries a proper Windows version stamp (visible in Properties → Details), which is what Windows Installer uses to decide whether to replace a file. Without it, it could decide to keep what was already there.
A failed install no longer looks like a successful one. If the installer can't complete, it now tells you on screen with the error code, writes a full log to %TEMP%\ROAR-install-<version>.log, and points you at it — instead of closing quietly as though everything worked.
The build no longer discards what the next upgrade needs. Packaging deleted the previous version's installer package, which Windows needs in order to remove the old version during an upgrade. Without it the upgrade failed and rolled back. Both the current and previous packages are now kept.
Upgrading
If you are on v0.34.0 or earlier and the update seems to do nothing, the reliable fix is to uninstall ROAR from Settings → Apps first, then run this installer. Your data is safe either way — settings, licence, history, snippets and vocabulary live outside the program folder and are never touched by install or uninstall.
Your existing licence and any trial in progress carry over untouched.
Full changelog: https://github.com/xhan145/roar/blob/main/CHANGELOG.md
ROAR v0.35.0 — 14-day Full-Feature Trial
Try every ROAR feature for 14 days
Open Settings → About and click Start 14-Day Trial. Every Pro and Developer feature unlocks — smart formatting, snippets, Code Mode, app profiles, per-app formatting and language, project vocabulary, vocabulary suggestions, history filters and tags, Flow automations, multi-output routing, and meeting capture.
No account. No payment card. No subscription. No internet. The trial is calculated and stored entirely on your machine, and it never starts on its own — installing, launching, or updating ROAR does nothing until you click the button.
When it ends, ROAR Core is still yours
Local dictation, multilingual transcription, streaming preview, history, privacy controls, basic vocabulary, and Scratch That keep working, free, forever.
Nothing you created is deleted or converted. Your snippets, profiles, tags, vocabulary, and settings sit exactly where you left them and light back up the moment a licence is activated. You get one calm notification when the trial ends — never during dictation, and never again after that.
A licence always wins
Buying at any point — during the trial or long after it ended — takes effect immediately in the running app, with no restart. Clearing history, deleting audio, or resetting settings can never restart, extend, or shorten a trial, and a normal update preserves it.
Honest about the clock
If your system clock moves backward, ROAR says so plainly, keeps Core available, and offers Try Again — no accusations. Offline trial enforcement is best-effort and documented as such in docs/TRIAL_ARCHITECTURE.md; it is not, and is never advertised as, DRM.
Privacy
The trial record holds only trial bookkeeping — never transcripts, audio, history, clipboard, vocabulary, or window titles. It is stored separately from all of your dictation data, and diagnostics show dates and states only, never key material.
Also in this release
Licences signed by ROAR's subordinate fulfilment key now activate offline exactly like founder-signed ones, so purchases can be delivered automatically. The founder signing key never leaves the owner's machine.
Install: download ROAR-Setup-0.35.0.exe below and run it. Windows 10/11, ~870 MB (the speech models ship inside — ROAR never uploads your audio).
1030 automated tests green.
ROAR v0.34.0 — Live transcript window + per-app routing
Two features that make ROAR feel like a workspace, not just a hotkey — plus your new fob button.
Live transcript window
Tray → Live transcript… (or right-click the fob) opens a compact, chat-style window showing everything ROAR transcribes as it happens — your dictations and meeting-capture lines interleaved. Filter All / Dictation / Capture, search, keep it always on top for calls and lectures, copy any line, or export the view to a text file. It reads your local history only — the window is physically read-only (SQLite read-only mode; it cannot write a byte) — and if history is off it says so honestly. Free in every edition.
Per-app routing profiles
Pin any output route on or off per application: "notes off in Slack" while notes is globally on, "notes always on in Word". Unpinned routes keep following your live toggles; typing into the focused app is never affected. Rules follow the app your dictation was aimed at — not whatever window grabs focus while transcription runs. Managed under Settings → Flow. (ROAR Developer)
Also
- The fob now wears its proper button artwork (with the drawn dot as fallback).
- Notes-route failures now pause notes routing everywhere — including pinned rules — with one honest notification instead of a toast per dictation; retoggling notes or changing the path retries.
- Settings is honest when a per-app rule is saved on a tier where it can't apply yet.
Install: download ROAR-Setup-0.34.0.exe below and run it (exit ROAR from the tray first). Fully local — no account, no telemetry, no subscription. The speech model downloads once on first run.
ROAR v0.33.0 — the mic fob
The dictation pill is now a fob — just like ROAR Android.
The mic fob
- Always there. Idle, ROAR shows a small round mic button you can put anywhere on any monitor. While you dictate, it expands into the familiar waveform pill right where it sits.
- Tap to talk. Tap the fob to start hands-free dictation; tap again to stop and type. Same behavior as double-tapping your hotkey — hotkeys still work exactly as before.
- Drag anywhere. ROAR remembers the position across restarts. If it was parked on a monitor that's now unplugged, it safely returns to bottom-center.
- Right-click for quick actions: Scratch that · Read selected text · Open Settings · Hide fob.
- Never steals your focus. Clicking the fob doesn't take keyboard focus from the app you're dictating into (
WS_EX_NOACTIVATE) — so your words always land where you were working. If Windows refuses the style, the fob turns display-only instead of risking typing into the wrong window.
Don't want it? Turn it off from the fob's own menu, the tray's "Show mic fob", or Settings → Voice & Mic — you get exactly the previous pill-only behavior back.
Also in this build
Everything from v0.32.0 (ROAR Flow automations & routing, live transcript view, meeting capture, OSC show control, Translate mode) — see the v0.32.0 notes if you're upgrading from v0.30.0.
Install: download ROAR-Setup-0.33.0.exe below and run it (exit ROAR from the tray first if it's running). Fully local — no account, no telemetry, no subscription. The speech model downloads once on first run.
ROAR v0.32.0 — Flow: automate it, route it, capture it
ROAR now does more than type what you say — it acts on it, routes it, and listens for you.
ROAR Flow — say it, ROAR does it
- Voice automations: say a trigger phrase at the start of a dictation and ROAR opens an app, opens a URL, presses a hotkey chord, inserts a snippet, speaks a reply, or copies text. Rules live in Settings → Settings → Flow and never fire mid-sentence.
- Multi-channel routing: one dictation can also go to the clipboard, a timestamped local notes file, and/or be spoken back — on top of normal typing, never instead of it. Toggle by voice ("roar route notes on", "roar routes off") or the tray Flow routes menu. Extra routes always start OFF, and a failing route can never block typing.
- Trusted rules: script, webhook, and OSC actions run only on rules you explicitly mark trusted, with a toast naming the rule when one fires.
Live transcript + meeting capture
- Live view: History gains a Live toggle — a chat-style local view that refreshes every 2 seconds — plus an Export button that writes the current view to a timestamped text file.
- Meeting capture (system audio): a tray toggle transcribes whatever your PC is playing — calls, meetings, videos — into local history via WASAPI loopback. No bot joins your call, nothing leaves your machine, and it never types into apps or touches your clipboard. Recording people may require their consent where you live.
Show control (OSC)
Trusted Flow rules can send OSC cues to lighting desks, QLab, Resolume, or TouchDesigner — "lights up" fires your cue. MIDI is on the roadmap.
Translate mode
Settings → Transcription → Translate to English: dictate in French, Spanish, Mandarin — ROAR types English. Whisper's built-in translate task, fully local, free in every edition.
Privacy, unchanged
Everything runs on your machine. The notes file is opt-in at a path you choose. The only possible outbound calls are a webhook or OSC datagram inside a rule you created and marked trusted — pinned by the network-hygiene test suite. No account, no telemetry, no subscription.
Install: download ROAR-Setup-0.32.0.exe below and run it (exit ROAR from the tray first if it's running). Settings → About → "Check for updates" now points here.
Editions: Core is free forever. Flow automations, routing, and meeting capture are ROAR Pro ($19 once); scripted/webhook/OSC actions are ROAR Developer ($29 once). Translate mode and the Live view are free. Gates remain display-only in this release.
ROAR v0.30.0 — Point & Speak, now working everywhere
Point & Speak: highlight anything, Ctrl + Right-click, hear it.
Select text in any app — your browser, a PDF, an editor, chat — and Ctrl + Right-click to have ROAR read it aloud on the spot. One gesture, no menu. The same gesture stops it. Everything stays on your machine.
Turn it on in Settings → Read Aloud → Point & Speak.
What changed since v0.26
The gesture shipped in v0.26 but often did nothing in browsers. Three separate causes, all fixed:
- The context menu was stealing the copy. The gesture click is now consumed by ROAR, so no menu opens to take focus mid-capture. Plain right-click is untouched everywhere.
- The gesture's own Ctrl key broke the copy. ROAR now waits for you to let go before copying, instead of sending a malformed Ctrl+C.
- The voice wasn't ready. With the gesture enabled, the voice now loads in the background at startup. Existing installs were hit hardest — their settings still carried
preload: falsefrom an older version, so the first gesture waited on a cold load and silently gave up.
Also: real accessibility access for Chromium apps (Chrome, Edge, Opera) that expose text only when asked; the clipboard is preserved in full — including images and copied files; and failures now say what actually happened instead of "this window isn't usable".
Verified end to end against a real Opera selection.
Install
Run the setup and follow the prompts. Your settings, history, licence and vocabulary carry over.
Requires: Windows 10/11 · Full detail: CHANGELOG.md
ROAR v0.26.0 — Point & Speak (Ctrl+Right-click to hear any text)
Point & Speak: highlight it, Ctrl + Right-click it, hear it.
Select text in any application — browser, PDF, editor, chat — and Ctrl + Right-click to have ROAR read it aloud instantly. One gesture, no menu, no extra click. The same gesture while it's speaking stops it.
- Any app, one pass — uses ROAR's privacy-hardened selection capture and the local Kokoro voice. Nothing leaves your machine.
- Plain right-click is untouched — ROAR only observes the modifier'd click and always passes it through, so every app's own menu still works exactly as before.
- Off by default — turn it on in Settings → Read Aloud → Point & Speak (requires Read Aloud to be enabled).
- Windows can't expose selections in administrator-elevated windows; those give a calm "no selection" notice.
Everything from v0.25.0 (Teach ROAR corrections, Read Aloud with GPU) carries over, as do your settings, history and licence.
Requires: Windows 10/11 · Full detail: CHANGELOG.md
ROAR v0.25.0 — teach it the words it mishears
Teach ROAR the words it keeps getting wrong.
Every voice has a few words that come out wrong every single time — a name, a bit of jargon, the way you say something. Now you fix it once and ROAR remembers.
Teach ROAR (new)
- From your history — press Teach ROAR on any past dictation, fix the word right there, and confirm. ROAR works out which word changed; you never retype the sentence.
- Or from Settings → Dictionary → Corrections — type what it heard → what you meant.
- It fixes and prevents. The correction rewrites that word in everything you dictate from then on, and what you meant is added to ROAR's vocabulary — so the recognizer is more likely to hear it right in the first place. That's the difference between patching the output and actually understanding you better.
- Your corrections stay on your machine, and they're free in every edition — accuracy for your own voice shouldn't be a paid tier.
Corrections only touch the misheard word: your punctuation, spacing and sentence casing are left exactly as they were.
Also in this release
Everything from v0.24.0, including Read Aloud — local, offline text-to-speech that's GPU-accelerated on NVIDIA hardware. Your settings, history, licence and vocabulary all carry over.
Install
Run the setup and follow the prompts. Requires: Windows 10/11.
Full detail: CHANGELOG.md
ROAR v0.24.0 — Read Aloud (local, GPU-accelerated speech)
Read Aloud comes to ROAR. ROAR can now speak — select or type text and have it read back in a natural local voice (Kokoro-82M), entirely on your own machine. No cloud, no account, nothing leaves your computer.
New: Read Aloud (Kokoro TTS)
- Fully local & private — synthesis runs in an isolated worker; ROAR never sends your text or audio anywhere. Read Aloud is off until you enable it, and Core dictation is unchanged.
- Optional download — to keep the base install lean, the voice model and its runtime are a one-time setup from Settings → Read Aloud, not bundled in this installer.
- GPU accelerated — with an NVIDIA GPU, setup installs CUDA and speech is ~10× faster (about a third of a second per sentence on an RTX 4060). CPU works everywhere. The model is preloaded and kept warm, and a one-time warm-up keeps the first request quick.
Everything else
Same fast, private local dictation as v0.23.0 — CUDA/Vulkan GPU paths, snippets, speech cleanup, per-app profiles, offline licensing. Your settings, history, and licence carry over.
Install
Run the setup and follow the prompts. Requires: Windows 10/11.
The model + runtime for Read Aloud are set up from inside the app after install; the GPU download needs a few GB of free disk.
Full detail: CHANGELOG.md