What changed, and when.

0.1.282026-09-06

• Windows: fixed HushQuill refusing to start with "The code execution cannot proceed because MSVCP140_1.dll was not found" on PCs that do not have the Microsoft Visual C++ runtime installed. The runtime files now ship inside HushQuill, so a fresh Windows install works without downloading anything else first. If you saw this error, install this version over the old one.

0.1.272026-08-07

• Fixed a licence check that could leave Pro features unlocked after a licence had already been withdrawn. If the licence server no longer recognised this device, HushQuill noticed but did not record it, so the paid features kept working for up to 30 days while the Licence panel showed Free.
• The Licence panel now explains itself when the server turns a key down, instead of quietly dropping to the Free plan. If your key stops working on this device you will be told, and asked to re-enter it.
• The update check no longer says "You're up to date" when there is simply no build published for your platform yet. It now says so plainly, so a missing Windows or Mac release is visible instead of looking like everything is current.
• Snippets: on the Free plan, when a snippet trigger would have fired, HushQuill now tells you it was pasted as spoken. The message existed but never reached the voice bar, so the feature looked broken rather than locked.
• Saying only a rewrite instruction, such as "make it formal", with nothing to rewrite yet now explains why those words were typed out instead of applied.
• Very short bursts of silence no longer type a stray word. Speech models invent filler like "you" when they are handed silence, and a short enough clip could slip past the silence check. Real short replies such as "Yes" are unaffected.
• The voice bar now identifies itself as HushQuill in the Windows task switcher instead of showing a placeholder name.

0.1.262026-08-05

• Fixed the Licence panel showing the Free plan after a restart even though your key was still saved. The panel could ask for its status before the app had finished starting, and then quietly kept showing Free for the rest of the session. This was the main reason a licence looked like it had been wiped.
• HushQuill now runs as a single instance. Starting it again brings the existing window to the front instead of launching a second copy. Two copies could overwrite each other's settings, including the licence.
• The summaries beside each collapsed section in Settings (Speech engine, Activation, AI Formatting) now match your saved settings as soon as the app opens. They could show startup defaults instead, so AI Formatting could read "Off" while it was actually on.
• A beta licence now shows the same expiry date wherever you are. It was worked out in your own time zone, so anyone behind UTC saw the day before the real one.

0.1.252026-08-04

• Licence activation errors now show the licence server's actual reason, such as "this key has reached the activation limit", instead of a generic "network: http status: 400".
• Re-entering a licence key on a machine where it is already active no longer uses up another device slot. Before this, every re-entry counted as a new device, so a key could hit its device limit from a single machine.
• Entering a key that belongs to a different product is now refused before anything is activated, so the mistake no longer risks using up one of that key's device slots.
• Your settings, including the licence, can no longer be lost when an update or shutdown closes HushQuill at the wrong moment. This was one of the causes of the app asking for a licence key again.

0.1.242026-08-04

• Beta tester licence keys now activate. Earlier builds rejected them with "This key belongs to a different product" because the app only recognised the Pro licence. A beta key unlocks every Pro feature on up to 3 devices for 90 days, and the Licence section shows the exact date it runs out.
• A licence with an expiry date now stops working on that date even when the machine is offline. Before this, an expired key could keep its features for up to 30 days if the app could not reach the licence server.

0.1.232026-08-03

• Fixed HushQuill closing instantly when you started speaking with the Whisper engine selected. On processors without AVX-512 (most AMD Ryzen laptops, and Intel chips outside the Xeon and 11th-gen-plus lines) the app was built with instructions those processors cannot run, so the first Whisper transcription crashed it. Parakeet and the cloud engine were never affected. This also applies to version 0.1.21, so if Whisper has never worked for you, this is why.
• Fixed the Licence section still showing "Free" beside its heading after you activated a Pro key. The section itself was correct when opened; only the summary line was stale.
• Fixed the same stale-summary problem on every other collapsed section. The engine, language, dictation mode and AI Formatting summaries all kept showing whatever they were set to when Settings opened, instead of your current choice.

0.1.222026-08-03

• Local transcription is substantially faster. HushQuill was making the speech model process a full 30 seconds of audio for every take, however short. Now it only processes what you actually said. Every language and every local engine benefits, most of all on laptops without a fast processor.
• Fixed HushQuill using too many processor threads, which was slowing transcription down rather than speeding it up. On an 8-core machine this alone roughly halved the time a take spends decoding.
• New "Higher accuracy" option for languages other than English. It uses a larger speech model that makes about half as many mistakes on Urdu, Hindi, Arabic and other non-European languages, in exchange for slower transcription. Free, optional, one-time download of about 190 MB.
• Free covers unlimited dictation in every language, all three local engines, cleanup and your personal dictionary. Pro adds unlimited AI formatting, the cloud engine, snippets, dictation history and your own rewrite styles. One payment, three devices, no subscription at any price.
• The cloud engine (Pro) is the most accurate option HushQuill offers, noticeably better than any local model for Urdu and Hindi. You bring your own free API key.
• New: dictation history (Pro). Your last 500 takes are searchable in Settings, so you can find something you dictated hours ago without saying it again. Stored on your device only: never uploaded, never synced.
• New: your own rewrite styles (Pro). Redefine what "email" or "chat" means to HushQuill in your own words instead of using the built-in wording.
• Settings is shorter and clearer. The Pro features sit together in one section you can fold away, the engine notes say plainly which engine is actually transcribing and why, and the cloud key field is marked and disabled until Pro rather than accepting a key that would not be used.

0.1.212026-08-02

• The Engine buttons now show which engine actually transcribes your language: an orange ring marks Auto's pick, or the engine HushQuill switches to when your selection cannot handle the chosen language.
• Fixed dictation getting stuck: text sometimes stayed on the voice bar and never reached your app, and the only way out was closing and reopening the bar. The bar now clears itself after every take and tells you what happened.
• When a paste cannot land, your words are never lost. HushQuill puts them on the clipboard and says so, whether the target is the desktop, an app running as administrator, or HushQuill's own Settings window. Ctrl+V or the bar's ⏎ button puts them where you want.
• Dictation can no longer wedge itself. A stuck transcription, a slow cloud request, or an unexpected error now recovers on its own instead of leaving the hotkey silently dead until restart.
• Stray words like "Thanks", "Yeah" and "Music" no longer appear in the middle of your text after you pause while speaking. These come from the speech model inventing sign-offs over silence, and they are now removed for each pause segment, not just at the very end.
• More filler words are cleaned up (ah, er, eh, mm, hm and friends) when cleanup is on.
• AI formatting now triggers when you expect it to. Spoken instructions work without a pause before them ("send the report Friday turn that into a formal email"), survive a trailing "thanks", and understand "convert that into bullets" and "rewrite that as" as well.
• Say an instruction on its own ("make it formal") right after a dictation and HushQuill rewrites what you just dictated.
• AI formatting is more reliable: the local model is warmed up while you speak so the first rewrite of the day no longer times out, and if the local model is unavailable HushQuill falls back to Groq when you have a key saved. When formatting genuinely cannot run, the bar says so instead of silently pasting unformatted text.
• AI formatting is now English-only and the whole card hides when another language is selected, rather than offering a switch that quietly does nothing.
• The Settings window now shows how many free AI rewrites you have left today, on both the AI Formatting and Licence cards, and the Upgrade to Pro button stands out as a proper orange button, with a glow once you have used the day's rewrites up.
• App styles are explained: each style says what it does, invalid lines are reported instead of silently dropped, and a Restore defaults button brings back the shipped list.
• Settings layout tidied: Activation now sits directly under Speech engine.

0.1.202026-07-30

• Click the voice bar to open Settings, and click it again to put Settings away. Dragging the bar to a new spot still works exactly as before.
• Fresh installs now start in English instead of auto-detect, which keeps the fast local engine in play from the first dictation. Your language is still one pick away in Settings, and existing settings are untouched.
• Fixed Settings sometimes opening with everything blank (language on Auto-detect, hotkeys showing only dots) instead of your saved choices. Saving from that state could overwrite your real settings. If settings ever genuinely fail to load now, the window says so outright rather than quietly showing you placeholders.

0.1.192026-07-24

• New: a full user guide inside the app. Click "Guide" at the top of Settings and it opens in its own window, so you can keep it open while you change things.
• The guide explains every feature in detail: both hotkeys, the voice bar, the three dictation modes, how the Auto engine picks for you, languages, AI formatting and spoken instructions, cleanup and your personal dictionary, snippets, voice commands, recovering text that did not land, microphone selection, and a troubleshooting section.
• It ships inside the app, so it works with no internet and always matches the version you have installed.

0.1.182026-07-24

• New: AI formatting. When you turn it on, HushQuill rewrites what you dictate to suit the app you are typing into: an email reads like an email, a chat message stays short, a code editor gets terse technical wording. Off by default; turn it on in Settings.
• You can now end a dictation with a spoken instruction and HushQuill will apply it and remove it from your text. Say "make that formal", "turn that into bullet points", or "format it as an email" at the end of what you are saying.
• Formatting runs on your own machine through Ollama by default, so your words never leave your computer. You can switch it to Groq in Settings if you prefer speed over local processing.
• If formatting is unavailable or fails for any reason, you still get your dictation exactly as before. It never blocks or loses what you said.
• Apps you have not mapped to a style, and dictations without a spoken instruction, skip formatting entirely, so normal dictation stays as fast as it was.
• Fixed: a spoken instruction after Urdu, Arabic, or Persian dictation was ignored, because those languages end a clause with their own comma and full stop.

0.1.172026-07-24

• New "Auto" engine, now the default: HushQuill picks the fastest speech engine for your language and your computer, so you get the best speed without changing any settings.
• Languages the fast local engine cannot handle (such as Urdu and Hindi) now switch automatically to Groq (when you have added a key) or the multilingual model, instead of quietly producing poor results.
• On older or slower computers, English dictation now uses a lighter, faster model that can be several times quicker than before.
• Faster Whisper transcription on Apple Silicon Macs, which now use the GPU.
• Settings: Language now sits above Engine, because your language guides the automatic engine choice. You can still pick a specific engine at any time.
• Mac: fixed microphone and accessibility permissions being lost after an update, so typing into other apps keeps working across new versions.

0.1.162026-07-23

• Added a "What's new" link in the header so the release notes are easy to find (not just by clicking the version number).
• HushQuill now greets you once after an update or a fresh install with a note pointing to what changed.

0.1.152026-07-23

• New "What's new" panel: click the version number in Settings to see every release and what changed.
• After an update installs, HushQuill shows a short "Updated to..." note so you know what's new.
• When an update is available, its release notes now show before you install it.

0.1.142026-07-23

• Up to 9x faster transcription with Whisper models. Urdu and other languages are now usable in real time on a typical laptop.
• Faster model benchmarking under the hood, so future speed regressions get caught before release.

0.1.132026-07-23

• Fixed a crash on launch that affected many AMD and older Intel machines.
• Better Urdu and Arabic-script dictation: text cleanup no longer garbles right-to-left text, and Arabic punctuation is handled correctly.
• The floating bar now displays right-to-left languages properly.
• The Fast (tiny) model is now English-only: the multilingual tiny model produced garbled output, so other languages always use the standard model.

0.1.122026-07-22

• Your dictated text is never lost. If no text box is focused when you finish speaking, the text parks on your clipboard and the bar tells you.
• New insert button on the bar: click into any text box and tap it to type your last dictation again.
• Changing a hotkey now takes effect reliably, even if you close Settings mid-change.

0.1.112026-07-22

• Model downloads now show live progress (percent and MB) instead of a static message.

0.1.102026-07-17

• The floating bar has clean edges (removed the faint halo around it).
• Fixed hotkey rebinding: the capture field now receives your key combo instead of sitting on "press keys" forever.

0.1.92026-07-17

• New Fast (tiny) model option for Whisper: about twice as fast, recommended on older machines.
• Your dictated words no longer appear inside the HushQuill window, only in the app you are typing into.
• Saved Groq API keys now show a "Saved" badge so you know you will not be asked again.
• Whisper no longer appends phantom sign-offs like "Thank you." after silence.
• First macOS build (universal, Intel + Apple Silicon).

0.1.72026-07-14

• Pick your microphone in Settings.
• New dictation modes: Toggle, Continuous (types each pause, keeps listening), and Push-to-talk.

0.1.62026-07-13

• Hands-free auto-stop: dictation types itself after a pause of silence.
• The bar is now persistent and mute-able, with a clickable mic button and its own hotkey.
• Start at login option.
• Whisper is now genuinely multilingual with a language picker (100+ languages).
• Settings window redesigned.

0.1.32026-07-12

• HushQuill now updates itself: automatic check on launch plus a "Check for updates" button in Settings.

0.1.22026-07-12

• New Groq cloud engine option (bring your own free API key), fastest transcription available.
• Snippets: say a trigger phrase, it expands into your saved template.

0.1.12026-07-11

• First Windows release: hold-hotkey dictation into any app, floating voice bar, fast local Parakeet engine, Whisper option, cleanup rules, personal dictionary, and voice commands.