What's new.

v1.2.1Latest

Jul 21, 2026

Pre-release hardening

• Password and other sensitive fields (PINs, card numbers) are now excluded from refining on every path — selection, keyboard shortcut, and detection alike.
• The Settings “Test connection” button now respects cloud consent: no request carries your key until the consent toggle is on.
• The in-page overlay got lighter on every site: its UI now mounts only when you actually select text, and background chatter during voice recording was cut sharply.
• Keyboard fairness: Alt+Shift+R passes through untouched on sites where you've turned RefineIQ off.
• Privacy policy refreshed to spell out the uninstall page's campaign tags, the connection test, and the sensitive-field exclusions.
• Requires Chrome 113 or newer (also the WebGPU baseline) — the store now enforces this.

v1.2.0

Jul 16, 2026

Sharper voice default, per-site control

• Dictation now defaults to the multilingual "Max accuracy" voice model — the most accurate on-device option. Lighter tiers stay selectable in Settings.
• New: turn RefineIQ off for specific sites under Settings → Sites, right from the options page.
• Reliability: transcripts can no longer be lost if you click away mid-transcription, concurrent refines no longer collide, and long drafts are never replaced by a truncated copy.
• Short drafts now expand slightly instead of ballooning with invented specifics — a measured rewrite-quality pass on the on-device model.
• Download progress is honest end to end: no more bars stuck at 100% on a fresh install, no saw-toothing percentages.
• This website launched — guide, blog, help, and the hosted privacy policy live here now.
• If you ever uninstall, an optional one-click feedback page helps us improve — still no accounts, no tracking, ever.

v1.1.0

Jul 5, 2026

Voice everywhere, no dead ends

• Voice Draft now works right on the page: focus any prompt field and a small Voice pill appears (or press Alt+Shift+V) — speak, and the transcript lands at your cursor, pre-selected so Refine is one tap away.
• A live waveform and a growing transcript preview show it's hearing you; recording stops by hand or hands-free after a few seconds of silence.
• Setup became a simple two-step checklist: the refine model, plus an optional Voice Draft step that handles the mic permission and speech-model download in one flow.
• Hit Refine before the model is set up? The download now happens right where you are — with progress — and your refine finishes automatically.
• Model pickers with live download progress moved onto the popup home, and the extension got a cleaner icon.