Voiceinput VoiceInput v0.73.0 ← Back
CHANGELOG

Every release creates value for users.

We don't pile features — we only ship what we believe earns its place.

v0.73.0 2026-06-01

Faster + more accurate translation

  • Cloud recognition / AI cleanup is faster: connection reuse makes text appear sooner after release in most cases.
  • Translation / bilingual is more accurate: proper nouns (company / product names) are recognized better.
  • Fixed the release date shown in the Update panel.
v0.72.1 2026-05-31

Added Simplified Chinese as a translation target

  • Added Simplified Chinese as a translation target: now you can speak a foreign language and translate straight into Chinese — quick-pick "简" from the menu bar.
v0.72.0 2026-05-31

New AI Translation + more accurate AI cleanup

  • New AI Translation: speak and get the translation directly, or original + translation side by side, 50+ languages (switch "Tidy / Translate / Bilingual" from the menu bar).
  • AI cleanup is more accurate: fixed the occasional case where it answered your question instead of just tidying what you said.
v0.71.1 2026-05-26

Small polish

  • Small polish.
v0.71.0 2026-05-26

Cloud speech recognition starts up faster

  • Cloud speech recognition starts up faster — pressing right Option begins recognition almost instantly.
  • Fixed occasional response stalls.
v0.70.1 2026-05-25

Auto-update prompt is harder to miss

  • Auto-update prompt is harder to miss: new banner at the bottom of the sidebar surfaces pending updates at a glance.
  • Auto-update is on by default after install — no confirmation popup.
v0.70.0 2026-05-24

AI cleanup back to generation-leap fast + onboarding upgrade

  • AI cleanup speed massively improved — release-to-text feels generation-leap fast again.
  • New-version updates are now harder to miss: after 24h the banner turns red, after 48h it auto-restarts to finish the update.
  • Onboarding redesigned: full-size keyboard with three-phase animation (press → speak → release & text appears) so you know which key to press at first glance.
  • Dashboard permissions: tap the "Needs approval" pill on the Accessibility row for an instant one-click fix.
v0.69.1 2026-05-21

Fix AI cleanup connectivity on certain networks

  • Fixed an issue where AI cleanup could not connect in certain network configurations.
v0.69.0 2026-05-20

AI cleanup noticeably faster + zero-setup + all-new onboarding

  • AI cleanup is noticeably faster — release the key and polished text appears almost instantly.
  • AI cleanup works out of the box, no configuration needed.
  • All-new onboarding flow so first-time users can get up and running quickly.
  • Long-sentence direct insert is more reliable and produces more complete results.
  • Recording status indicator is clearer and more polished.
  • Proper-noun recognition is more accurate.
  • Fixed sign-in failures on certain network configurations.
v0.68.1 2026-05-15

Fix first-launch "Apple cannot verify" dialog

  • Fixed an occasional "Apple cannot verify VoiceInput" dialog on first launch of v0.68.0. Other features identical to v0.68.0.
v0.68.0 2026-05-14

AI tidy speed boost + company / tool / Chinese-idiom mishearing auto-fix + chat & email tone preserved

  • AI tidy responds noticeably faster — the whole "release-and-see-text" loop now feels much snappier.
  • AI company / tool name recognition improved: Anthropic / OpenAI / DeepSeek / Cursor / TypeLess and other common tech terms get heard right far more often.
  • Common Chinese idioms now auto-corrected when misheard: 「头痛医头脚痛医脚」 「实事求是」 「不胫而走」 and similar fixed phrases are restored to their proper form.
  • Chat & email contexts now preserve your spoken tone: still drops filler words and adds punctuation, but no longer forces your message into stiff business prose.
v0.67.0 2026-05-13

Maintenance release

  • Maintenance release. No changes to daily usage.
v0.66.4 2026-05-13

Local-insert duplication fixed + AI tidy more reliable

  • Fixed local-insert occasional text duplication: when ASR re-identified an entire utterance mid-sentence, the cursor would show duplicated content. Now fixed.
  • AI tidy more accurate: company names / tech terms / homophone corrections improved; output stays closer to your original phrasing.
  • AI tidy learning loop: after you manually correct an AI-tidy output, the system remembers the proper-noun mapping and gets it right next time automatically.
v0.66.1 2026-05-11

Menubar picker sync + streaming focus check fix

  • Menubar ASR engine switch synced to drop "Local Streaming": v0.66.0 updated the settings page but missed the menubar Picker — fixed here.
  • Streaming AI tidy focus stability fix: prevents early characters from being dropped when focus is switching back after release.
v0.66.0 2026-05-11

Streaming AI tidy + local engine unified to SenseVoice

  • Streaming AI tidy: hold the AI-tidy key, speak, release — text now streams typewriter-style into your cursor position with the first character appearing in under 0.5s, instead of "wait 2-3 seconds then see the whole paragraph".
  • Local engine unified to SenseVoice: the "Local Streaming Paraformer" engine has been retired. The more accurate "Local SenseVoice" becomes the sole local engine; existing Paraformer users auto-migrate on next launch.
v0.65.2 2026-05-08

More empty-recordings caught upfront

  • More empty-recordings caught upfront: if your mic captures nothing during a recording shorter than 3 seconds, releasing the key now immediately prompts you to check mic permission/volume, instead of round-tripping to cloud ASR for an empty result.
v0.65.1 2026-05-08

Steadier mainland-China cloud connectivity

  • All client ↔ cloud traffic (cloud ASR, AI tidy, quota checks) moved to a custom domain api.voiceinput.app, bypassing the prior default subdomain that was throttled or DNS-poisoned on some mainland-China networks.
v0.65.0 2026-05-07

Major cloud ASR stability + silent-mic guard + auto-jump to engine settings on missing model

  • Major cloud ASR stability: when cloud ASR is overloaded, VoiceInput now auto-falls back to a local engine and writes the result to clipboard, instead of retrying the same failing path.
  • Silent-mic guard: if the mic captures no sound during a recording, you get a clear "no sound from mic — check permission/volume" instead of a silent empty result.
  • Auto-jump to engine settings on missing model: picking a local engine without its model downloaded now auto-opens Settings → Engines so you can grab it in one click.
v0.64.0 2026-05-07

Fix Bluetooth headset second-sentence eats first + occasional stale-clipboard paste + tighter AI tidy

  • Fix Bluetooth headset second-sentence eats the first: the first sentence used to get deleted and rewritten when you started speaking the second. This version adds a sentence-boundary detector — the first sentence stays put, and the new one appends after it.
  • Fix occasional paste of previous clipboard content: between two consecutive AI tidies, the second paste sometimes pasted the previous result. This version fixes the underlying clipboard restore timing.
  • AI tidy more robust: when the model output drifts too far from the source / drops a key phrase (e.g. splitting "first principles" into two pieces) / contains repeated chunks, VoiceInput auto-falls back to raw audio. Weird tidy results never reach your cursor.
v0.63.0 2026-05-02

Silent background download + dashboard restart prompt

  • Silent background download + dashboard restart prompt: when a new version is detected, VoiceInput now downloads the DMG silently in the background. Next time you open the main window you'll see a purple card "✨ Version vX ready · Restart now to update" — one click applies the update instantly, no download wait and no Sparkle modal.
  • Fix Pro users seeing "AI · not configured" in the left sidebar status strip when using cloud AI. v0.59.2 had marked this fixed but only updated the right "System Status" panel — the left sidebar status row was missed.
v0.62.0 2026-05-02

Stale-version nudge + AI tidy stability + livelier weekly digest

  • Stale-version nudge on launch: if your version is 3+ minors behind the current release, you get one gentle "new version available" prompt that opens Sparkle directly (snoozed for 24h after dismiss).
  • Livelier "This week" digest: previously the weekly digest kept generating nearly identical content. Now it picks fresh items from a larger candidate pool each time — headlines and trend text no longer repeat.
  • AI tidy stability: long-text tidies no longer fall back to the heuristic on transient timeout — results are more complete.
v0.61.1 2026-05-02

Fix "Source App" stats duplication

  • Fix "Source App" stats duplication: previously the same app was counted as multiple sources because of different window titles (e.g., "WeChat" + "WeChat · WeChat (chat)"), making dashboard rankings inaccurate. This version unifies them by app name, with automatic migration of old entries.
v0.61.0 2026-05-02

Clearer local model missing message + auto-retry AI tidy

  • Clearer local model missing message: previously when "local streaming" was selected but the model wasn't fully downloaded, every hotkey press just said "ASR connection failed" — confusing. Now it says "Local X model not downloaded — go to Settings → Engine to re-download" directly.
  • AI tidy retries once on network failure: occasional network glitches used to make a recording "wasted"; now AI tidy auto-retries once before falling back to raw, recovering most cases.
v0.60.1 2026-05-01

Fix double-clicking app icon doing nothing

  • Fix double-click app icon doing nothing: after install, double-clicking the app now opens the main window directly instead of relying on the menu bar icon. Fixes macOS 26 sporadic menu-bar-icon-not-showing issue that left users with no way in.
v0.60.0 2026-05-01

Flash routing wrap-up + AI transparency + undo window

  • "AI changed N spots" badge: every record in Data Center now shows a small badge with how many things AI rewrote, click to see exact from→to list — no more black box.
  • Undo AI tidy 30s: menu bar shows an "Undo" button for 30 seconds after each AI tidy. One click puts the raw text back on the clipboard so you can ⌘V to replace the tidied version.
  • Smarter to-do recognition: "urgent" / "must" auto-sets high priority; "every day" / "every Monday" / "weekdays" auto-sets recurrence; negation guard ("I don't want to do X" no longer mis-classified as a to-do).
v0.59.2 2026-04-30

Fix AI status display + internal stability improvements

  • Fix: When a Pro user selected "VoiceInput Cloud AI", the sidebar still showed "AI · Not configured". Now correctly reflects cloud AI as configured.
  • Internal stability improvements: the app now auto-captures crashes and key-flow errors so we can find and fix issues faster.
v0.59.1 2026-04-29

Fix "todo routing + note app double-write" bug

  • Double-write bug fix: previously when you said "Remind me to call Wang at 3pm", a Reminder got created ✓ but flomo / Apple Notes also saved a copy. This release: when intent routes to Reminders or Calendar, the note-app distribution is skipped — content only goes to the right place.
  • Cleaner feedback toast: when routed to Reminders, the toast shows "✅ Added to Reminders"; for Calendar, "📅 Added to Calendar". No more "distributing to flomo" misleading message.
v0.59.0 2026-04-29

"Knows-you" trio: personal vault · window context · usage profile

  • Critical fix for "第一性原理 / 第一要义" mis-split: previously these fixed phrases got wrongly split into 1.2.3 list items because "第一" was treated as a numbered marker without context. This release adds strict-monotonic-ascending number detection — repeated fixed phrases no longer get split.
  • "My Info" personal vault (Settings → Preferences): fill in email / company / position / English name / phone / address — then say "my email" or "my company" and it auto-expands to the stored value. Sensitive fields (phone / address) encrypted in macOS Keychain, all local, never uploaded.
  • Window context awareness: beyond app name, we now read the frontmost window title (e.g. "Email to Zhang Peng - Lark Mail") so AI tidy knows who you're replying to, outputting more context-fitting text.
  • Personal usage profile: client builds "Top apps / Top contacts / Top topics" summary from your last 90 days locally, then feeds AI tidy prompts so the LLM knows your long-term focus areas. Pure local, user can disable.
v0.58.4 2026-04-29

Todo + Calendar NER · clean titles · precise time parsing

  • Todo routing now NER-structured: previously the whole sentence was stuffed into the title, now we extract time and a clean title separately. E.g. "Remind me to call boss Wang at 10am tomorrow" → Reminder title="Call boss Wang" due=tomorrow 10:00 (system alarm fires on time).
  • Schedule → macOS Calendar (new): when recognized as an event (e.g. "Meeting with client at 2pm next Wed"), auto-creates a calendar event with precise startDate / endDate (default 1 hour + 15-min advance alarm). Toggle in Settings → Sync to Notes "Smart routing → Schedule" (off by default).
  • Chinese date-time NER: supports today / tomorrow / day-after / next week X / weekday names / X month X day / morning / afternoon / evening + clock time / Chinese numerals (一二三...) up to 20.
v0.58.3 2026-04-29

Flash v2 · todos auto-route to Reminders · AI tidy quality guard

  • Flash todo detection (experimental): when you say "remind me to call mom / don't forget to buy milk / I need to write the report tomorrow", the flash gets recognized as a todo and written to macOS Reminders. Toggle in Settings → Sync to Notes (off by default). Pure client-side heuristic, no network.
  • AI tidy hallucination guard: occasional model output like "Here is the tidied:" / "Sure, let me tidy:" gets stripped automatically. Garbage output (< 3 chars / all punctuation) falls back to raw transcript.
  • List detection: added markers for spoken enumerations like "one is / another is / one more is", "first / second / third (一来/二来/三来)", "on one hand / on the other hand".
v0.58.2 2026-04-28

Settings switched to top horizontal tabs (matches mock)

  • Settings page layout rebuilt: was "secondary nav sidebar + detail pane" (macOS-utility style), now "top horizontal tabs + full-width detail" — matches the modern style of the main window. Tab selected state has a purple underline + purple text.
  • Detail area gets full width, each SettingsSectionCard renders with more breathing room.
v0.58.1 2026-04-28

ESC actually works + settings overhauled to lower the bar

  • ESC cancel actually works now: previous version only covered Flash + global monitor missed events when our own app was frontmost. Now global+local dual monitors fully cover. Any recording (Direct / AI tidy / Flash) can be cancelled with ESC — no ASR, no insert, no DB write.
  • Settings 4 tabs each get a purple intro card: one-line "what this section controls + new-user recommendation".
  • "Sync to Notes" reorganized: icons grid (3 columns) at top showing all 9+ supported note apps (flomo / Apple Notes / Bear / Drafts / Tana / Yuque / Joplin / ...), configured destinations below.
  • "Tidy" AI source upgraded to radio-card style: Soundecho Cloud vs Bring-Your-Own API Key — clearly two-choice with purple selected border.
v0.58.0 2026-04-28

Main window overhaul · know what to do next when you open the app

  • New "Next step" card at the top of the main window: smartly shows newbie guide / recently unlocked badge / streak about to break / how far to the next badge / unused features. Opens the app no longer staring at dead statistics.
  • Fifth stat card replaced with badge progress: shows unlocked / 30 + current streak + distance to next badge.
  • Sidebar items now show subtitles. "Data Center" renamed "My Records", "Imprint" renamed "My Footprint".
  • Settings 4 tabs renamed: Voice → Recording / AI Tidy → Tidy / Distribution → Sync to Notes / System → Preferences. No more engineer jargon.
  • Flash brand color unified: yellow ⚡ → purple ⚡.
  • Flash ESC cancel: press Esc to immediately cancel an accidental flash key trigger.
v0.57.7 2026-04-28

List transition words stripped cleanly

  • Fix "first goal is X" extracted as "goal is X": previously list items still carried transition phrases like "goal/thing/reason/idea/step/key/point is …". This release strips those too — each list item starts with the actual content from the first character.
v0.57.6 2026-04-28

Audio storage 8× smaller + forever-retention option

  • Audio files switched from PCM to AAC: 1 minute went from ~2MB to ~240KB. Local-only compression, never uploaded. "Re-transcribe from raw audio" experience is unchanged.
  • Old recordings auto-migrate: 1 minute after launch, old .pcm files compress to .m4a in the background. Failures retain the original for next-attempt retry — your data is never lost.
  • New "forever" retention option: pick "永久 / forever" in the retention dropdown to disable auto-cleanup entirely.
v0.57.5 2026-04-28

List numbering fixed (1. 2. 3. instead of all 1.)

  • Critical bug fix: the previous version rendered all three list items as "1." (should be 1. / 2. / 3.) because the renderer assumed it was outputting to a markdown engine that auto-renumbers. But your input field is a plain text box that does not. This release writes real numbers directly.
v0.57.4 2026-04-28

Lists only when truly enumerating + 3-second hard wall

  • Better list detection: previously "I went to eat, then worked, then went home" would also be wrongly split into a 1/2/3 list — now only clearly enumerative scenarios ("first / second / third") become lists; narration and chat stay flat.
  • Fix "第一个/第二个" (the n-th) extraction: "第一个是 X" used to come out as "2. 个是 X" (with stray character) — now stripped cleanly.
  • Fix "首先 + 第N" mixed structure: "首先...第一是...第二是..." now treats "首先..." as paragraph intro, 1/2/3 as the real list.
  • AI tidy 3-second hard wall: anything over 3s falls back to raw insert, no more pointless waiting.
v0.57.3 2026-04-28

Speed back to 2-3s + lists actually render as lists

  • Speed restored: key-release to result is back to 2-3s (the previous release's forced 5s+ wait was a wrong-direction call and has been reverted).
  • List scenarios truly become lists: when you say "first / second / third", "one is / two is / three is", "first / then / finally", the result renders as a real list — not a flat paragraph.
  • Paragraph scenarios stay flat; no structure is forced where there isn't any.
v0.57.2 2026-04-28

No more 6-second wait + hotkey order matches keyboard layout

  • No more freeze on key release: in the previous version, if you had an audio-direct model configured, you could wait up to 6s for nothing. Now you get whichever finishes first by 2.5s — never held back artificially.
  • Hotkey choices now read left-to-right across the keyboard: the picker order for Direct-insert / AI-tidy / Flash-capture follows actual key positions, so you don't have to scan back and forth.
v0.57.1 2026-04-29

List items are now real takeaways, not raw transcript

  • Before: "First it's X, second it's Y" came out as "1. it's X / 2. it's Y" — barely tidied. This release pushes the AI to actually distill the core point (verb-phrase / noun-concept), so list items are cleaner and easier to scan.
v0.57.0 2026-04-29

Fix: AI tidy structure was being preempted

  • If you had an audio-direct model configured (Qwen / Gemini), it would typically return text in 1-2s and beat the structured AI tidy to the result — meaning lists / paragraphs / headings from AI tidy were never reaching you. This release gives AI tidy a 2.5s priority window: if it returns within that window, its structured result wins; otherwise the audio-direct fallback kicks in.
  • Badge unlocks now sync to the cloud, so My Footprint progress will stay consistent across devices.
v0.56.0 2026-04-29

AI tidy now understands intent first, then formats

  • AI tidy logic upgraded: instead of mechanically applying rules, it now first identifies the intent of what you said (narration / enumeration / argument / comparison / summary / venting), then formats accordingly. Lists become lists, paragraphs stay as paragraphs, sectioned content gets headings — no more structure for structure's sake.
  • A more stable Chinese model is being rolled out gradually on the backend; tidy quality continues to improve.
v0.55.2 2026-04-29

Lists actually render as lists now

  • When the model returned structure metadata in an inconsistent shape, the client used to fall back to a flat paragraph. This release adds a client-side rescue: phrases like "first... second... third..." or "to begin with... then... finally..." are now split into a list even when the model mislabels them.
  • Tidy response window widened from 2.2s to 6s — fewer in-flight requests get cancelled when running behind a proxy.
v0.55.1 2026-04-29

Fix: AI tidy was broken behind an HTTP proxy

  • When running behind an HTTP proxy (VPN / Clash, etc.), the streaming AI tidy could come back empty, leaving you with the raw speech-to-text instead of the polished result. This release adds an automatic fallback to a non-stream call when the stream returns nothing, so you reliably get the cleaned-up text.
v0.55.0 2026-04-29

List rendering polish + new entry to AI Memory

  • List rendering polish: even when the model only labelled segments loosely, the result is now preserved as separate lines instead of being collapsed back into a single paragraph.
  • New entry on the Data Center: a "Try natural-language search" prompt that takes you straight to AI Memory — search with phrases like "things from last week's meeting" or "what I said on WeChat".
v0.54.1 2026-04-29

Fix: list structuring was silently dropped

  • AI was already detecting that a passage should become a list, but the client side was discarding that structure — you ended up seeing a flat paragraph at the cursor. Now real lists / headings / paragraphs make it through.
v0.54.0 2026-04-28

Smarter list detection

  • Say "first... second... third..." or "to begin with... then... finally..." and the result automatically becomes a clean bullet list, no longer flattened into a single paragraph.
  • Meeting cues are picked up faster too: words like "decided / action item / owner / next step" trigger the meeting structure (decisions / action items / follow-ups) without needing a long monologue.
  • Typography polish: cleaned up several Chinese strings across Settings, Feedback, and share cards that were using half-width punctuation by mistake.
v0.53.0 2026-04-28

Short-press feedback + clearer failure diagnostics

  • Pressing the hotkey too briefly no longer silently fails: you now get a clear "Hold a bit longer" toast.
  • Recording failures now show more specific reasons in the data center, making it easier to report issues and fix them.
v0.52.0 2026-04-28

AI tidy speed boost

  • Major AI tidy speed boost: text appears the moment you release the key — feels nearly as instant as raw insert.
  • The preview bubble now shows progress earlier during recording — no more "I pressed the key, nothing's happening".
v0.51.0 2026-04-27

My Footprint · Three-dimensional badge collection

  • New "My Footprint" page in the main window: 30 book-plate style badges across three dimensions (Words / Time / Streak), each with 10 levels and metallic tiers (Bronze / Silver / Gold / Royal). Tap any badge for details + copy share text. Existing users have full history backfilled automatically.
  • Words map to literary classics (Tao Te Ching / Analects / Outcry / Gatsby / Jane Eyre / Crime & Punishment / Dream of Red Chamber / Les Misérables / Shakespeare). Time maps to public-speaking benchmarks (hourglass / sunrise / dipper / armillary / compass / galaxy / ouroboros). Streak maps to habit milestones (seed / sprout / flame / full moon / torch / pine / phoenix / scales / eternal).
  • New optional Dock icon: defaults to menu-bar-only (stays minimal), opt in via Onboarding or Settings → System → Display.
  • Sign-in emails adapt to scenario: first sign-in / returning user / invited friend.
v0.50.5 2026-04-26

Stability fixes bundle

  • Uninstalling and reinstalling no longer randomly pops the keychain password dialog — opening Settings used to eager-read three Volcano-engine keychain items (even if you don't use Volcano). Now they're read lazily only when you actually open the Volcano config card.
  • License activation is more resilient: in rare cases activation returned "not found" instantly; now it auto-retries once after 800ms, catching transient network blips.
  • Opening the main window now puts the app icon in the Dock for ⌘Tab switching; the icon auto-hides when you close the main window.
  • When a paid license is stacked onto an existing trial license, the whole license is now correctly promoted to paid — the UI no longer mistakenly shows "Pro trial".
v0.50.1 2026-04-26

One-click fix for "permission granted but app does nothing"

  • New "Reset permissions" button (in onboarding and permission guide) for the recurring "I toggled it on in System Settings but the app still does nothing, only restarting helps" issue. Clears macOS's stale TCC records under the hood — re-authorize once and you're back.
  • Dashboard system status: if Accessibility looks granted but the app actually can't use it, the row now auto-shows "Tap to fix".
v0.50.0 2026-04-26

Dashboard + Pro card visual rebuild

  • Dashboard rebuilt to spec: dark mode aligns with voiceinput.app, light mode same-hue mapping. Single sentence summary "spoke X min, saved ≈ Y min of typing today" + 4 stat tiles + system status panel + weekly bar chart + source apps ranking.
  • Settings → Account Pro card rebuilt: violet gradient + three mono columns (LICENSE KEY / expiry / device count), device cell now shows the real "2 / 5" count.
  • System status panel adds five rows (ASR engine / AI model / Accessibility / Mic / Hotkey) with data-source descriptions + three-color capsule (green ok / yellow needs config / gray neutral).
v0.49.0 2026-04-26

🎁 1 month of Pro on first sign-in

  • First-time email sign-in unlocks 1 month of Pro on the house — unlimited cloud speech recognition + unlimited AI tidy. No credit card, no auto-renew.
  • Earlier users who signed in but never paid get the same trial backfilled.
  • Settings → Account Pro card now shows trial countdown ("X days left") and expiry date.
  • Fixed two v0.48 account-card details: the "show" link was unclickable, and the device cell showed "·" which looked like activation failed.
v0.47.0 2026-04-26

Referrals + visual rebuild

  • Referrals are live: Settings → Account shows your invite code and short link. When a friend signs in via your link and installs the app, you both get 1 month of Pro.
  • Visual rebuild: now follows system light / dark mode, aligned with voiceinput.app brand (deep indigo + violet accent).
  • Pro activated card new look: violet gradient + JetBrains Mono license key (click to reveal, auto-mask after 30s).
  • Dashboard system status panel adopts the capsule three-color rule: green = running / amber = needs setup / gray = neutral info (e.g. hotkey).
  • Sign-in page rebrands to dark style: violet gradient halo + "声" logo.
v0.46.2 2026-04-26

Overview status reflects your real configuration

  • The Dashboard "System status" panel is now accurate: ASR engine shows whichever you actually selected (Local SenseVoice / Apple / Volcano / Cloud); AI model shows the model your "Tidy recording" task is actually bound to (e.g. deepseek-v4-flash) instead of a fixed one.
v0.46.1 2026-04-26

Auto-activate Pro after sign-in + clearer model status

  • Pro users are activated automatically after email sign-in — no need to paste a license key.
  • System status panel now shows "Cloud AI · Pro" for Pro users; BYOK users see their configured model name plus a configuration check.
v0.46.0 2026-04-26

Optional email sign-in

  • Settings → Account adds an "Email Sign-in (optional)" entry. Sign in to link your Macs together — works on the free plan too.
  • Click the email link and you're sent right back to VoiceInput. If the deep link is blocked, you can still copy the session token and paste it back into the app.
  • Sign-in email and verify page are now bilingual (Chinese / English), following your system language.
v0.44.9 2026-04-25

Accounts + Pro subscriptions + cross-device licenses

  • Email-based account login. See all activated Macs at voiceinput.app/account, with one-click remote revoke for old devices.
  • Pro subscription unlocks unlimited cloud ASR + AI tidy. Menu bar shows "Pro · Unlimited" instead of quotas.
  • "Bring Your Own Key" path is committed to be free forever, never gated behind Pro.
  • Homepage adds a "Three ways to use it" comparison card: 100% Local · Bring Your Own Key · Shengyi Cloud.
  • Cloud ASR reliability greatly improved. When recognition fails, the app silently re-runs cloud transcription in the background and writes the result to your clipboard — just hit ⌘V to recover.
v0.43.1 2026-04-25

In-app language switcher

  • Settings → System now has a "界面语言 / Language" toggle: Follow system / 中文 / English. Restart in one click after switching. Now Chinese-system users can manually use the English UI too.
v0.43.0 2026-04-26

English version is live + AI tidy prompts follow your system language

  • Every screen, menu, setting, toast and error message is fully translated. Switches automatically with your macOS system language.
  • AI tidy now uses a dedicated English prompt when your system is in English — tuned for English homophones, punctuation, and removing fillers ("um", "uh", "you know").
  • voiceinput.app root URL now redirects based on your browser language; the top-right toggle is still there if you want to override.
v0.42.3 2026-04-25

Mini pill sits further from the text

  • The streaming mini pill now sits about 1 cm below your cursor (was ~3 mm), so it no longer crowds the text you are typing.
v0.42.2 2026-04-25

Cloud speech recognition usage now counts real recording time

  • Cloud speech recognition usage is now billed against your actual recording length. Previously the "waiting for server final result" round-trip (5-30 s per call) was silently added to your quota, piling up to fake minutes/hours.
v0.42.1 2026-04-25

"Did it actually hear me?" anxiety, addressed

  • AI tidy now also shows what you are saying live: the mini pill auto-expands a second line with the live transcript (direct insert is unchanged since the text streams straight into your input field).
  • Flash capture confirmation got much stronger: when you finish a flash, the center bubble does not immediately vanish — it morphs into a big green checkmark + "Saved · delivered to flomo · Apple Notes" and stays for 2.5 seconds. No way you miss it now.
v0.42.0 2026-04-25

Flash delivery feedback + unified direct/AI tidy interaction

  • Flash delivery feedback: as soon as you finish a flash, you see "⚡ Saved · delivering to flomo · Apple Notes…", and the toast auto-upgrades to "✅ Delivered to …" once dispatch completes (or tells you which destination failed).
  • Flash list delivery footprint: each flash card bottom now shows a tiny icon row of which apps received it.
  • Direct insert and AI tidy now share the same caret-anchored mini indicator. If tidy takes longer than 1.5 s, a partial preview line slides out below to ease the wait. A small green tick blinks at completion.
  • When AI tidy fails it now silently falls back to the raw transcript instead of nagging you with a toast.
v0.41.4 2026-04-25

Indicator next to your cursor + direct insert is steadier

  • The streaming-mode floating indicator shrunk from 200×44 to a 116×26 mini pill, and auto-anchors just below the input field you are typing into (falls back to bottom-center semi-transparent if it cannot locate the caret).
  • Fixed cloud streaming + direct insert occasionally dropping text — when the ASR was slow with no partials, the final text could be silently lost. Now double-protected.
v0.41.3 2026-04-25

Bubble gets out of the way

  • When streaming insert is active, the floating bubble now moves to the bottom-right corner and becomes semi-transparent, so it stops covering the text you are typing into.
v0.41.2 2026-04-25

Quotas split, no more double display

  • Menu bar splits the cloud quotas into two clearly-labeled rows: "cloud speech recognition" (monthly, 60 min) and "cloud AI tidy" (daily, 50 calls). The old single "cloud quota exhausted" line made people think the whole cloud died — they are independent quotas.
  • When cloud speech recognition is exhausted, we now explicitly say AI tidy is unaffected.
  • When streaming insert is active, the floating bubble no longer repeats the recognized text — your text already streams into the input field, the bubble just keeps the waveform + "recording" state.
v0.41.1 2026-04-25

Streaming insert everywhere, quota fixed

  • Real-time streaming insert now works in every input field by default — only terminals, password managers, and System Settings are skipped for safety.
  • Fixed the menu bar wrongly showing "quota exhausted" when plenty of cloud credit remained.
v0.41.0 2026-04-24

Real-time streaming insert

  • While dictating with the direct-insert key, text now streams into the cursor in real time — no more waiting until you release.
  • Added a "Real-time streaming insert (experimental)" toggle under Settings → Voice.
v0.40.0 2026-04-24

Main window and settings, slimmed down

  • Sidebar trimmed from 4 items to 3: My Records / Flash / AI Memory.
  • Settings collapsed from 11 tabs to 4: Voice / AI Tidy / Distribution / System.
  • AI tidy daily quota copy fixed from 30 to 50.
  • Menu bar quota badge is now visible across all ASR engines.
v0.39.0 2026-04-24

Cloud quota at a glance

  • The menu bar now shows your remaining cloud ASR quota for the month.
  • When close to the limit it suggests switching to local streaming; when fully exhausted it switches automatically.
  • AI tidy daily quota raised from 30 to 50.
v0.38.4 2026-04-24

More accurate, and faster

  • Even without AI tidy, "Anthropic ... Claude" and "Cursor API" style phrases are auto-corrected.
  • AI tidy wait time shortened — you get the final text sooner after you stop speaking.
v0.38.3 2026-04-24

Visible usage, easy selection

  • Every AI tidy call now appears in the usage log — see which model was used and how long it took.
  • Audio tidy model is now a dropdown — no more typing model names.
  • Added auto-correction for common misheard brand names — even without AI, names come out right.
v0.38.2 2026-04-24

See every tidy in action

  • A small toast after each AI tidy tells you which path was used and how long it took.
  • Every entry in Data Center shows how it was tidied, so you know exactly what happened.
  • Qwen auto-upgraded to the latest version, more stable.
v0.38.1 2026-04-24

Flash can go to more notes apps

  • Flash delivery adds three new destinations: Tana, Yuque, and Drafts.
  • Add-destination grid expanded to 9 icons so you can see everything at a glance.
v0.38.0 2026-04-24

Let AI hear what you actually said

  • Recordings can be sent directly to a multimodal model for listen + tidy — proper nouns like Anthropic / Claude / Cursor get recognized far better.
  • Text still appears as fast as before — the accurate version quietly replaces it in the background.
  • New "Audio Tidy" section in Settings → AI Connection, just pick a model and paste your key.
v0.37.2 2026-04-24

Bear delivery + more hotkeys

  • Fixed the Apple Notes same-text-three-times bug.
  • Flash capture can now be sent to Bear.
  • Hotkey options expanded with easier-to-reach keys: left Shift / right Shift / right Control / Caps Lock.
v0.37.0 2026-04-24

AI tidy stops adding filler

  • AI tidy no longer tacks on "In summary..." or "Hope this helps" — just the clean text you meant.
  • No more stray emoji or Markdown symbols.
  • New "Flash delivery": your flash captures can auto-push to flomo / Apple Notes / any custom endpoint.
v0.36.0 2026-04-24

Introducing "Flash Capture"

  • Tap the flash key to start, tap again to stop. No need to open any app — just speak, and it's saved to Data Center.
  • First launch now proactively prompts for all three permissions so you don't have to dig through System Settings.
v0.35.0 2026-04-24

Local true-streaming

  • New local streaming engine: fully offline, text appears as you speak.
v0.33.2 2026-04-23

All-new onboarding

  • Opens a 4-step onboarding after install: welcome → grant 3 permissions → try it once → start using. Permission checkboxes update automatically — no need to click refresh.
  • Menu bar adds a "Show onboarding again" entry; revisit anytime.
  • DMG install background redone — no more broken-glyph boxes.
v0.32.1 2026-04-23

Stability fix

  • Fixed the occasional self-quit crashes in recent builds.
v0.32.0 2026-04-23

AI tidy no longer gets stuck

  • AI tidy no longer hangs with no text appearing. Even on slow networks, within 3 seconds you will see the raw transcript inserted with a gentle toast telling you the tidy timed out.
  • The "recognition returned empty" problem is significantly reduced.
v0.31.0 2026-04-22

A few small AI tidy quirks

  • Fixed the occasional "stuck on recognizing" that would not move on.
  • Sentence-final periods are no longer quietly swallowed after AI tidy.
  • No more four-em-dash-in-a-row ugly artifacts.
v0.30.0 2026-04-21

Menu bar information architecture

  • Menu bar gains a one-tap speech engine switcher: Cloud / Local / Apple / Self-configured.
  • The Copy / Re-tidy / Re-transcribe buttons under Recent recognition are now larger and more visible.
  • "Recognition failed" now only shows the single latest failure instead of a growing backlog.
  • Removed the obsolete "Output mode" switch (its purpose is now covered by the two hotkeys).
v0.29.0 2026-04-21

Cloud streaming reliability lifted to local levels

  • Fixed several intermittent failure modes in cloud streaming ASR ("nothing came back", "stuck for 5 seconds"). Cloud success rate now matches local.
v0.28.0 2026-04-20

Press-to-respond is back, plus AI tidy feedback

  • First-token latency is back under 100 ms (down from 0.5–1.5 s in v0.27.3) — the press-and-it-shows-up feel is restored.
  • The overlay flips from "Recording" to "Transcribing…" the moment you start speaking, so the system feels alive instead of frozen.
  • AI-tidied records now show 👍 / 👎 in Data Center. Stored locally, never uploaded — used internally to iterate on the prompt.
v0.27.3 2026-04-20

Real-volume waveform + recover failed recognitions

  • Overlay waveform is now driven by your actual microphone volume — it jumps the moment you start talking, instead of running a fake time-based animation.
  • A new banner at the top of the menu bar popover lets you re-recognize any "failed" recording from the last 24 hours with one tap — no need to dig through Data Center.
  • Turned off the v0.25.0 ASR pre-warm pool. Pre-warmed connections were getting silently dropped, causing "I spoke but nothing came out". Back to per-press connection: +300-800 ms first-token latency, but recordings work 100% of the time.
v0.27.2 2026-04-19

Hotkey options fixed: Left Control + Fn

  • Removed the broken "Right Control" option (Macs do not actually have that key). Added "Left Control ⌃", which every Mac has.
  • Added "Fn 🌐" as a hotkey option — well-positioned on MacBook and Magic Keyboard.
  • Users who picked "Right Control" in v0.27.1 are auto-migrated to "Left Control" on upgrade.
v0.27.1 2026-04-19

Empty recognitions no longer get lost

  • When recognition returns empty, a "Failed" entry is now saved with the original audio attached. Right-click "Re-transcribe" in Data Center to recover, no need to repeat yourself.
  • New hotkey option "Right Control ⌃" (rarely conflicts with system shortcuts). Right Command now shows a gentle "easy to mis-press" warning.
v0.27.0 2026-04-19

See your recordings folder + cleaner hotkey config

  • Settings → Privacy → Raw Recording Retention adds a "Show in Finder" button — always know where your local audio lives.
  • Hotkey settings simplified: removed the redundant three-way "Output mode" picker, replaced with a more focused "AI tidy behavior" config.
v0.26.1 2026-04-19

Dual-hotkey settings, now visible

  • Settings page now clearly displays both rows: 🎙 Direct insert + ✨ AI tidy. What you see is what you get.
v0.26.0 2026-04-19

Two keys, two scenarios

  • Right Option and Left Option each bind to one output mode.
  • Direct insert for chat (no AI, fastest). AI tidy for documents (cleanest output).
  • Train the muscle memory once — two clear scenarios.
v0.25.0 2026-04-19

ASR pre-warming, press and respond

  • The app pre-warms one ASR connection 3 seconds after launch.
  • First-press response latency drops from 300-800 ms to <50 ms. Auto re-warms after each recording.
v0.24.2 2026-04-19

Quick re-actions from the menu bar

  • "Re-transcribe" and "Re-tidy" buttons added to the Recent Recognition section in the menu bar popover — no need to open Data Center.
v0.24.0 2026-04-19

Local audio archive + one-click re-transcribe

  • Recordings are saved as PCM locally (default 7 days, configurable 0-30).
  • Right-click any history entry to "Re-transcribe from original audio" or "Re-tidy" — no need to repeat yourself when ASR cuts off or tidy disappoints.
  • Audio files never leave your Mac.
v0.23.3 2026-04-19

AI tidy quality lifted + truncation fix

  • Tidied output now adds a thin space between CJK and Latin, and auto-corrects brand casing (Anthropic / OpenAI / Kimi, etc.).
  • Punctuation present in the original transcript is preserved, never silently stripped.
  • Fixed an intermittent bug where the preview showed 50 characters but only 20 actually got inserted.
v0.22.1 2026-04-19

Tidy gets its own model config

  • The core flow — record → AI tidy → insert — now has its own dedicated config row.
  • Want Kimi for tidy and DeepSeek for category fixes? Both, side by side, no interference.
v0.22.0 2026-04-19

API account system: configure once, used everywhere

  • No more entering the same API key three times across the Fast / Standard / Reasoning tiers.
  • Accounts managed centrally; each tier just picks "which account + which model". Six dropdowns to configure everything.
  • Existing users auto-migrate with full data preserved.
v0.21.5 2026-04-19

Apple-notarized DMG: opens cleanly on any Mac

  • Starting this version, the DMG is signed with a Developer ID Application certificate, Apple-notarized, and stapled.
  • New users can double-click and run, no warnings.
  • Existing users see a one-time "Certificate upgrade" wizard on first launch — 30 seconds, then never again.
v0.21.4 2026-04-18 hotfix

Tag cloud actually clicks

  • Tap a tag in AI Memory to jump to the matching records. No more retry-tapping.
v0.21.3 2026-04-18

Sharper mixed-language recognition

  • Words like "Cursor, Kimi, API" stop getting misheard in Chinese context.
  • AI cleanup focuses on corrections — no more rewriting your tone.
v0.21.2 2026-04-18

Chinese typography, closer to handwriting

  • Half-width space between CJK and Latin reads softer than a full space.
  • Five new typography details: em-dash, quotes, ellipsis, and more.
v0.21.0 2026-04-18

Data Center and AI Memory split

  • Your raw records live in Data Center — searchable, filterable, exportable.
  • AI's interpretations, persona takes, and weekly portrait live in AI Memory.
v0.20.0 2026-04-17

AI Memory: personas + weekly portrait + quotes

  • Same line re-read by 7 personas: Boss, Friend, Business Coach, Musk, Jobs, Therapist, Editor.
  • Custom personas: name + emoji + tone — see your words through anyone you know.
  • Weekly Quotes: AI picks 3–5 resonant lines, each with an open follow-up.
  • Weekly Portrait: Big-5 based snapshot, presented on MBTI axes.
v0.19.1 2026-04-17

Your records are never lost again

  • Upgrades and config changes never wipe history.
  • Seven auto-backups on each launch. Uninstall takes local data with it.
v0.19.0 2026-04-17

Settings reorganized + one-click rule packs

  • Accuracy gets Rule Packs: Programming, Filler Words, AI Brands, Chinese Fixes, Business — tap to enable.
  • Settings grouped into Voice Input / AI & Data / System.
  • "AI Model" renamed to "AI Connection" — configure once, reused everywhere.
v0.18.0 2026-04-17

Overlay visual refresh

  • Overlay height follows content — no more large empty space.
  • During AI tidy, original text stays with a purple breathing glow.
  • Font size, padding, corner radius, shadow all lightened.
v0.17.0 2026-04-16

Live text preview + free cloud AI

  • Words appear in the indicator in real time as you speak.
  • AI tidy briefly shows a correction diff, then fades.
  • Free cloud AI cleanup: works out of the box, 30 times a day.
v0.14.0 2026-04-16

Typography engine + 1–2s end-to-end gain

  • Spaces, punctuation, case fixed locally — no network, no AI.
  • 45+ brand names auto-corrected: kimi → Kimi, github → GitHub.
  • Even without AI, voice input gets clean typography.
  • End-to-end latency: 3s → 1.5–2s.
v0.13.0 2026-04-15

Better AI cleanup + learning inbox

  • Pinyin disambiguation: homophone pairs no longer confused.
  • Keeps your spoken style — no more rewriting casual verbs.
  • Learning inbox in menu bar: AI extracts candidate rules from your edits.
v0.12.0 2026-04-15

Invisible cleanup: polished output, one shot

  • After you stop talking, wait 1–2s — the polished version lands in one paste.
  • Three tidy levels: minimal / natural / structured.
  • Target window locked — switching apps falls back to clipboard with a toast.
  • Double-tap Right Option to skip AI for one round.
v0.10.0 2026-04-14

AI architecture + learning loop

  • Built-in provider presets: Doubao, DeepSeek, Kimi, Qwen, OpenAI.
  • Learn from your edits — suggest new hotwords and rules.
v0.9.1 2026-04-14

Auto-update + perf fixes

  • Daily update checks, with release notes before install.
  • Scroll lag fix on large Data Center lists.
v0.9.0 2026-04-14

UI refactor

  • Fresh visual language: quiet precision.
  • Overview page becomes "Today".
  • Data Center becomes "Memory Stream".
v0.8.0 2026-04-14

Multi-engine + zero setup

  • Zero setup: no API key required on first run.
  • VoiceInput Cloud: 60 free Chinese ASR minutes per month.
  • Local SenseVoice: fully offline.
  • Apple native ASR: zero-config offline fallback.
v0.7.0 2026-04-14

AI output, simplified

  • Two modes: Direct Input, AI-Polished Input.
  • Both insert instantly; AI runs in background, zero perceived wait.
v0.5.1 2026-04-12

Latency crush

  • Press-and-speak: no cold-start lag.
  • Head-of-audio loss fixed.
  • System-level noise suppression on.
v0.5.0 2026-04-12

Accuracy pass

  • Three-layer accuracy: hotwords → local substitution → AI fix.
  • Custom hotword system for proper nouns.
v0.4.0 2026-04-12

AI Memory first pass

  • Auto-categorization, natural-language queries, one-line summaries.
  • Daily briefing, weekly insights.
v0.3.0 2026-04-11

Data Center

  • Auto-save voice input history.
  • Search, filter, favorite, export.
v0.2.0 2026-04-11

Core MVP

  • Global hotkey voice input: hold to speak, release to insert.
  • Real-time Chinese ASR.
  • Menu bar always-on with recording indicator.