Changelog

What's new.

Notable changes, roughly in reverse chronological order.

  1. v0.13.6 More reliable Codex assistance
    • Fixed Codex responses stay in the right order. Streamed text is now assembled in sequence before it is displayed or spoken, preventing scrambled words and response sections.
    • Changed The Codex model picker reflects your account. Available models now come directly from Codex, with a local cache to keep the picker useful if a refresh is temporarily unavailable.
  2. v0.13.5 Safer corrected dictation
    • Fixed AI Correction keeps prompt guidance out of your text. Terminology hints now stay separate from the dictated transcript, so dictionary expansions such as assistant prompts are less likely to pick up internal correction instructions.
    • Changed The Super Voice panel feels more polished. Drawer positioning, meeting rows, live notes, and feed presentation have calmer defaults.
    • Fixed Annual licenses recover more smoothly. Expired annual subscriptions now follow the grace-period path before access changes.
  3. v0.13.4 Sharper AI-corrected dictation
    • Changed AI Correction better understands spoken repairs. It is more careful with phrases like I mean, actually, no wait, and correction, so literal mentions stay intact while real corrections keep the words you intended.
    • Fixed Installed correction models stay local. Super Voice Mode now prefers usable local model snapshots before checking the network, avoiding surprise re-downloads when the Gemma correction model is already available.
    • Changed Rewrite results appear near where you are working. The rewrite HUD now opens closer to the active caret or pointer.
  4. v0.13.3 Cleaner meeting notes and snapshots
    • Changed Meeting review is calmer. Very short speaker fragments are treated as unknown, so review focuses on the voices with enough audio to be useful.
    • Fixed Meeting snapshots now render in Slate. Snapshot links in meeting notes display as local images instead of raw markdown.
    • Changed Meetings copy is clearer. Privacy and terms copy now better describes local meeting capture, snapshots, transcripts, and saved meeting files.
  5. v0.13.2 Better meeting speaker review
    • Changed Meetings are better at keeping speakers reviewable. Short one-off turns stay in the transcript, and speaker controls are easier to reason about on calls with similar voices.
    • Fixed Speaker review and Slate received polish and fixes. Remote speaker controls behave more predictably, and saved Slate files are easier to find.
  6. v0.13.1 Meetings Beta speaker review
    • Added Meetings can now keep speaker labels easier to review. Review unknown speakers after a call, name them, and keep the transcript readable while staying in control.
    • Added The Meetings library highlights calls that need review. New filters and flags make it easier to find meetings with unnamed speakers, local-speaker review, or unknown platforms.
    • Changed Meetings is now labeled Beta. The feature has moved beyond Preview, with richer metadata, cleaner review, and local-first controls.
    • Fixed Slate handles meeting files more safely. Meeting file controls now follow the file in view, multiple file tabs work again, and closing a meeting tab no longer leaves edits attached to the wrong note.
  7. v0.13.0 Vocabulary for AI-corrected dictation
    • Added Custom Vocabulary helps AI Correction spell the terms that matter to you. Add names, acronyms, product terms, medical terminology, or other specialized words, and Super Voice Mode will use them as local context when cleaning up dictation.
    • Changed Vocabulary works across short and long dictation. Relevant terms are shortlisted per dictation and passed to the local correction model as guidance, not blind replacements.
    • Changed Workspace now has a dedicated Vocabulary pane. Add one term, paste a comma-separated list, import a text file, or open the vocabulary file directly when you want full control.
  8. v0.12.14 Live Dictation UI improvements
    • Changed Live Dictation UI improvements.
    • Fixed Various bug fixes and quality-of-life improvements.
  9. v0.12.13 Sharper dictation capitalization
    • Fixed Dictation no longer adds stray capitals when you continue a sentence. Pause and keep talking and the next words stay lowercase as a continuation, while genuine sentence starts, names, and the word I stay capitalized.
    • Changed The Assistant settings tab is now called External Access. It groups what a connected outside agent is allowed to reach, with clearer labels.
  10. v0.12.12 Better meetings and note editing
    • Added Richer meeting notes with a live summary that leads with action items. Notes now capture attendees and source, and the summary surfaces who owns what as the meeting happens.
    • Added Edit your notes and save them straight back to your files. Unsaved changes are tracked and you're warned before you lose them.
    • Changed Connect more AI assistants, plus assorted polish and fixes.
  11. v0.12.11 Improvements and fixes
    • Fixed Background voice actions, dictation, and Super Voice improvements and fixes.
  12. v0.12.10 Better AI-corrected dictation
    • Changed AI Correction is better at understanding when you change your mind mid-sentence. Phrases like I mean, no wait, actually no, and scratch that are handled more reliably, so corrected dictation keeps the words you intended.
    • Changed Setup explains AI Correction more clearly. First-run onboarding now highlights local dictation cleanup, Launch at Login, and the on-device model that powers both corrected dictation and AI Writing.
  13. v0.12.9 Improvements
    • Changed Meeting, AI Correction, and Super Voice improvements and fixes.
  14. v0.12.8 Meetings feel more like files
    • Added Hermes Agent joins the assistant backends. Connect Super Voice Mode to a local Hermes Agent server, choose the Hermes profile you want, and keep turns grouped into real Hermes sessions.
    • Added Meeting snapshots land where you expect. Snaps taken during a meeting now save straight into that meeting's snapshots folder, with a quiet camera cue when the capture is saved.
    • Changed Local AI uses memory more carefully. Local MLX models stay warm when they help responsiveness and unload when idle, so speech, correction, and assistant work fight less over resources.
    • Changed Meeting controls are calmer. The Snap button now shows when a capture is in progress and avoids accidental rapid-fire screenshots. If you finish right after snapping, Super Voice Mode waits briefly so the snapshot can be included.
    • Changed Meeting notes stay file-native. Meeting entries open the real notes.md file, recent meetings follow the folders on disk, and title edits no longer disturb folders you have organized yourself.
    • Fixed Hermes replies no longer repeat themselves. Hermes Agent streaming now handles repeated or cumulative chunks without doubling the assistant's final message.
    • Fixed Meeting speaker labels are steadier. Super Voice Mode is more careful about separating linked app audio from the local room mic, so speakerphone bleed is less likely to appear as a phantom local speaker.
    • Fixed The Super Voice panel is steadier during live meetings. The panel avoids a layout loop that could make the app use too much CPU while meeting fields were updating.
  15. v0.12.7 Meeting notes get steadier
    • Added More reliable live meeting transcripts. Meeting capture now keeps a durable transcript flowing while you record, recovers more gracefully between chunks, and gives you clearer controls when a meeting needs to continue.
    • Changed Meeting notes behave more like real files. Notes and metadata now round-trip through the meeting's notes.md, so what you edit in Slate stays aligned with what is saved in your vault.
    • Changed The Super Voice panel resizes more naturally. The live panel is easier to resize, with a cleaner edge hit area and better filtering in the activity feed.
    • Fixed Cleaner download and usage numbers. Download analytics now separate people, bots, command-line downloads, scanner traffic, and tagged website links, so the dashboard reflects real usage more closely.
  16. v0.12.6 The Slate becomes an editor
    • Added Edit anything on the Slate. The Slate is now a true editor. Type anywhere in the document or in any vault file, with tables, diagrams, math, and images rendering in place as you write. Tables edit cell-by-cell with right-click row and column actions, undo works everywhere, and a reading mode locks the page when you just want to read.
    • Added Your vault, in the sidebar. Browse your vault as a folder tree: pin files, search, and drag to organize, including dragging files in from Finder. Switch vaults from the header, and visit your Characters, Meeting Templates, and Augments folders to edit them right on the Slate.
    • Added Resize images. Drag the corner handle to size an image in a document, and the size is saved in the file itself. Image files open fitted to the window, with pinch and zoom controls.
    • Added Name a meeting while it's happening. Type a title during the call and it carries through to the saved notes. Your title always wins over the calendar's.
    • Changed Meetings live in your vault now. Each meeting is saved as a folder of ordinary files (notes, transcript, and snapshots), so you can search them, sync them, and open them in any app. Existing meetings move over automatically, with a backup kept. Everything still happens on your Mac.
    • Fixed Your edits always land. Changes to open files are saved when you switch tabs, close the window, or quit. And if a save ever fails, your text stays on screen and retries instead of being lost.
  17. v0.12.5 Meetings
    • Added Meetings (Preview). Capture a call and get structured notes, fully on-device. Super Voice Mode transcribes the conversation, separates speakers, and writes the notes locally, so nothing leaves your Mac. Notes and the full transcript open side by side, with a live transcript in the Super Voice panel while you record. It can offer to take notes automatically when you join a call, and matches your calendar to fill in the title and attendee names. Export to Markdown or draft a follow-up email in a click. Available with Pro; turn it on or off anytime from the Meetings tab.
    • Added Meeting notes stay private by design. Notes are generated in a locked-down assistant mode that can only read and summarize (it can't run tools or take actions), and the transcript is treated as untrusted text, so something said on a call can never act as a command.
    • Changed Better computer use (Preview). The assistant is more capable inside real apps and websites: choosing from menus, handling picker fields, clicking the right controls, and checking each step actually worked before moving on.
  18. v0.12.4
    • Changed Computer text entry now uses the keyboard first. When the assistant fills editable fields, it now favors normal typing events and keeps Accessibility value setting available as an explicit compatibility route for apps that need it.
    • Fixed Draft replacement is safer. The assistant now refuses to overwrite non-empty fields unless replacement is explicit, verifies the focused field before clearing, and fails closed instead of appending into an uncertain composer.
    • Fixed Text-entry checks are stricter. Super Voice Mode no longer treats nearby page text as proof that a field was filled; it now requires readback from the actual target where possible.
  19. v0.12.3
    • Added Computer use is much more capable. The assistant is better at working inside real apps and websites: choosing from menus, handling picker fields, clicking the right controls, and checking that each step actually worked before moving on.
    • Added Two new voices. Adam and Orion join the built-in voice set for spoken replies.
    • Changed Custom characters are clearer and safer. Super Voice Mode now more clearly separates built-in characters from character files you can edit yourself, with stronger warnings when custom instructions can shape assistant behavior.
    • Fixed Setup windows behave better. Loading, onboarding, and What's New windows are more resistant to macOS sizing glitches.
    • Fixed Working-directory picker fixes. Choosing an assistant working folder is more reliable from both the Super Voice panel and Settings.
  20. v0.12.2
    • Added Full computer use (beta). Point your assistant at any app, any website, any browser: it can click buttons, fill in text, pick from dropdowns, scroll, and read what's on screen, quietly in the background. It never takes over your screen, moves your mouse, or steals your focus, and anything uncertain stops and asks you first.
    • Added Scrolling and dropdowns. The assistant can now scroll pages and pick from dropdown menus, and it verifies each selection actually took before moving on.
    • Added Control speech with the keys you already use. While the assistant is talking, tap your hotkey to skip ahead, hold it to pause and dictate, or double-tap to clear everything queued. There's a Skip button in the popover and panel too.
    • Added The assistant goes quiet when you talk. Starting a recording pauses spoken replies and resumes them after, so you're never dictating over the assistant.
    • Added See how much room the conversation has left. The assistant panel now shows how much of the conversation context has been used.
    • Changed 50 free Pro uses each week, now live. After your Pro trial ends, the free tier includes 50 Pro uses every week (assistant turns, Live Transcription, and spoken replies), resetting each Monday. Upgrade to Pro when you want them unlimited.
    • Changed Rebuilt computer control, safer by design. The tools your assistant uses to see and act on your screen were rebuilt from the ground up: every action checks it's still looking at the window and page it expects, your blocked apps and sites are enforced everywhere, and anything uncertain stops and asks you first.
    • Changed More reliable typing into busy apps. Keystrokes the assistant sends are now paced so fast-moving apps and web pages don't drop them.
    • Changed The panel opens on the right screen. On multi-display setups, the assistant panel now appears on the display where you're actually working.
    • Changed Spoken commentary follows your acknowledgment setting. One picker now controls how chatty the assistant is out loud, across all assistants.
    • Changed A leaner toolset for connected apps. External apps connected over MCP now see a focused set of tools instead of everything, and voice output from those apps only happens when explicitly requested.
    • Fixed The assistant panel stays on your monitor. On multi-display setups, the panel and side notch no longer jump to another screen mid-conversation.
    • Fixed Sturdier model downloads. A damaged download is now repaired and retried automatically instead of failing.
  21. v0.12.1
    • Changed 50 free Pro uses each week, now live. After your Pro trial ends, the free tier includes 50 Pro uses every week (assistant turns, Live Transcription, and spoken replies), resetting each Monday. Upgrade to Pro when you want them unlimited.
  22. v0.12.0
    • Added Full computer use (beta). Point your assistant at any app, any website, any browser: it can click buttons, fill in text, pick from dropdowns, scroll, and read what's on screen, quietly in the background. It never takes over your screen, moves your mouse, or steals your focus, and anything uncertain stops and asks you first.
    • Added Scrolling and dropdowns. The assistant can now scroll pages and pick from dropdown menus, and it verifies each selection actually took before moving on.
    • Added Control speech with the keys you already use. While the assistant is talking, tap your hotkey to skip ahead, hold it to pause and dictate, or double-tap to clear everything queued. There's a Skip button in the popover and panel too.
    • Added The assistant goes quiet when you talk. Starting a recording pauses spoken replies and resumes them after, so you're never dictating over the assistant.
    • Added See how much room the conversation has left. The assistant panel now shows how much of the conversation context has been used.
    • Changed Rebuilt computer control, safer by design. The tools your assistant uses to see and act on your screen were rebuilt from the ground up: every action checks it's still looking at the window and page it expects, your blocked apps and sites are enforced everywhere, and anything uncertain stops and asks you first.
    • Changed More reliable typing into busy apps. Keystrokes the assistant sends are now paced so fast-moving apps and web pages don't drop them.
    • Changed The panel opens on the right screen. On multi-display setups, the assistant panel now appears on the display where you're actually working.
    • Changed Spoken commentary follows your acknowledgment setting. One picker now controls how chatty the assistant is out loud, across all assistants.
    • Changed A leaner toolset for connected apps. External apps connected over MCP now see a focused set of tools instead of everything, and voice output from those apps only happens when explicitly requested.
    • Fixed The assistant panel stays on your monitor. On multi-display setups, the panel and side notch no longer jump to another screen mid-conversation.
    • Fixed Sturdier model downloads. A damaged download is now repaired and retried automatically instead of failing.
  23. v0.11.7
    • Added Two new assistants: Cursor and OpenCode. Pick Cursor or OpenCode as your voice assistant, right alongside Claude, Codex, and the local model. Each comes with its own model picker.
    • Added Your assistant asks before it acts. Tool actions now go through a quick approval prompt, so the assistant checks in before running something rather than doing it all on its own.
    • Added Iris, built in. Iris now ships with the app and is the default voice for spoken replies.
    • Added Mute for a session. Silence spoken replies for the current session in one tap, without touching your saved settings.
    • Changed Cleaner assistant setup. Regrouped the assistant settings and refined the new Cursor and OpenCode panes so choosing and configuring an assistant is simpler.
    • Changed A more polished live panel. A panel you drag stays put, your spoken replies are kept for replay, the voice meter follows playback, and the activity feed weaves your words and the assistant's replies together.
    • Fixed Smoother audio. Moved audio processing off the realtime path for cleaner capture during longer sessions.
    • Fixed Better reading of your apps. The assistant now reads window content reliably, including Electron apps, instead of cutting off early.
  24. v0.11.6
    • Added Weekly free Pro uses. After your Pro trial, the free tier now includes 50 Pro uses each week for Live Transcription, assistant turns, and spoken replies. Pro remains unlimited.
    • Added Working directory picker. Choose where your assistant works right from the Super Voice panel: a quick dropdown with your favorite and recent folders, and a star to pin the ones you use most. It's now a single setting every assistant shares.
    • Changed Codex keeps its train of thought. When Codex is your assistant, it now holds context reliably across a back-and-forth, so your follow-up questions stay in the same conversation.
    • Changed History moved into Settings. Your dictation and assistant history now lives in its own Settings pane with a cleaner timeline. Jump straight to it from the panel footer.
    • Changed A calmer Super Voice panel. Tidied up the live panel with repositioned footer controls and lighter styling for a cleaner look.
    • Fixed Smoother spoken replies. Starting a new request now cleanly takes over from whatever's being spoken, so replies no longer overlap.
  25. v0.11.5
    • Added History. A new History view gives you a timeline of your dictation and assistant activity, color-coded by source. Click any spoken line to replay it in place. Loads a recent window with "Load older" for going further back.
    • Added Playback controls. Pause, resume, and stop spoken replies right from the menu-bar popover. Pressing to talk now cleanly cuts off whatever's being spoken so you're never talking over the assistant.
    • Changed A live Super Voice panel. The assistant drawer is rebuilt as a single live panel that shows your dictation and the assistant's replies as they happen, color-coded by source. It peeks open on its own when you start talking and tucks away when you're done. On by default; the classic drawer stays available as a fallback.
    • Changed Cleaner assistant setup. Removed the "Recommended" tag on backends and a permissions warning that no longer applied, so choosing an assistant is less cluttered.
    • Fixed Spoken replies no longer talk over themselves. The assistant won't play its short acknowledgment while it's already speaking.
    • Fixed A sensible default voice. New installs fall back to a built-in voice when none is set, instead of staying silent.
    • Fixed Accurate model list. The Models pane no longer counts unrelated downloaded models toward your installed voices and LLMs.
  26. v0.11.4
    • Fixed Cleaner Slate replies. Your assistant's prose no longer picks up stray protocol tags. Only genuine document content lands in Slate, even when you mention things like <slate> mid-sentence.
  27. v0.11.3
    • Added Live Transcription. Double-tap the dictation key to start a hands-free session with live subtitles. Words stream to the screen in real time with AI corrections, great for emails, meeting notes, and longer writing.
    • Added A context-aware assistant. The voice assistant now reads the window you're looking at and knows which app you're in, and can act: drive Mail, Calendar, and Notes, open URLs, and remember your preferences across sessions.
    • Added What's New window. A curated feature-highlights window appears on first launch after an update, so you always know what changed. Also available from the app menu and the Dashboard Help menu.
  28. v0.11.2
    • Added Assistant activity at a glance. The Insights pane now shows an analytical summary of what your assistant did (which tools ran, the apps it touched most, how much data moved, and its success rate) instead of a raw log. Turn it on in Dashboard → Assistant → Insights.
    • Fixed Reliable app detection for connected assistants. Connected assistants now correctly identify the app you're focused on for every action, instead of occasionally acting on a stale one.
  29. v0.11.1
    • Added Smarter spoken replies. The assistant now speaks just its final answer aloud and keeps mid-task narration silent, so you hear the conclusion without the play-by-play.
    • Fixed More natural speech. Contractions, possessives, and stray symbols are handled correctly so spoken output sounds smoother.
  30. v0.11.0
    • Added Agentic Context (Beta). Voice Mode now gives a connected AI assistant a set of tools to see your screen and act on your behalf: read the window you're looking at, pull your highlighted selection, take a screenshot when text alone isn't enough, paste results back into your app, drive scriptable apps like Mail, Calendar, Notes, Messages, and Finder, open apps and URLs, and remember a few stable facts about you across turns. Connect once from Dashboard → Assistant → Tools.
    • Added Safer pasting. When the assistant pastes into your frontmost app, Voice Mode first checks that your focus is actually an editable field, so a stray paste can't land on a button or accidentally hit Send.
    • Added Personal preferences. Your assistant can remember small, stable facts about you (how you sign emails, naming preferences, who's on your team) kept in a file you own and can edit, and capped so it can't quietly balloon.
    • Added Audit trail. Every time a connected assistant reads from or acts on your Mac, Voice Mode logs a one-line entry and plays a soft tone, so you always have passive awareness of what's being touched. Visible live in the Insights pane.
    • Fixed License countdown. The trial countdown no longer shows "0 days left" after your trial has actually ended. It now correctly switches to the upgrade card.
  31. v0.10.9
    • Added Insights (Beta). A new pane reflects your dictation back to you: a voice signature that captures your pitch, cadence, and pace at a glance, a words-per-hour waveform across your day, the words and phrases you reach for most, and a quiet summary of what your AI corrections were actually doing (formatting, translating, rewriting tone, and so on).
    • Changed Your dictation history is now encrypted on disk. Every line is sealed before it touches storage, with the key kept in your Mac's Keychain. Existing history upgrades automatically on first launch, and nothing about how dictation feels changes.
  32. v0.10.8
    • Changed A warmer onboarding. Granting each permission now flashes green and chimes, with a calmer "all set" beat when the last one lands. The "Try it" step shows your dictation speed after each phrase, and finishing onboarding lands on a calm "You're set" screen with your visualizer, your hotkeys at a glance, and one concrete thing to try first.
    • Changed Something to watch while the AI model downloads. The download step plays a short looping demo of what's coming ("make this shorter," "translate to French," "rewrite in a friendly tone") so the wait isn't wasted.
    • Changed A clearer license shortcut. The sidebar shortcut now shows the action that fits where you are: "Upgrade to Pro" on the free plan, "Renew Pro" if an annual plan lapsed, and a calm countdown like "5 days of Pro left" during a trial that turns urgent in the final days.
  33. v0.10.7
    • Fixed The Codex assistant works again out of the box. A change in the Codex command-line tool left fresh installs unable to start it without manual setup. Voice Mode now drives it correctly, so the Codex and Gemini assistants just work after install.
    • Changed Codex is ready to go on new installs. The Codex assistant works the moment you press the hotkey, with no manual setup. You can adjust what it's allowed to do under Assistant → Codex Settings, and existing setups keep whatever you already configured.
  34. v0.10.6
    • Added A thank-you when your trial wraps. The first launch after a trial ends shows a short greeting: what stays free, what Pro unlocks, and a soft nudge toward the voice assistant.
    • Changed Pro features stay visible to free users. Slate, Assistant, Voice, Personas, and Augments now show in the sidebar with a lock curtain inside, instead of being hidden until you upgrade, so it's clear what Pro adds.
    • Changed Buttons stay vivid when another window is in front. Primary actions across the app used to fade to grey when the Dashboard wasn't the active window, especially in light mode. They now keep their accent color so they stay easy to find.
    • Fixed Trials now end correctly. An expired trial properly drops to the free plan at launch, instead of sitting on an active trial indefinitely.
  35. v0.10.5
    • Changed One-click Slate from its pane. The Slate pane now puts an "Open Slate" button right next to the pitch, so the demo is one click away.
  36. v0.10.4
    • Changed Update banner stays put. When a new version is out, a banner pins to the top of the Dashboard and stays there until you install. Closing the update window doesn't dismiss it anymore. The only way to make it go away short of installing is "Skip This Version."
    • Changed New versions reach you faster. Voice Mode now checks for updates every few hours while it's running, instead of once a day. Releases land on your machine the same day they ship, not the day after.
    • Fixed The update banner stops disappearing on you. A handful of edge cases in the update check could make the banner blink off after a few hours even when an update was actually waiting. It now stays lit from the moment a new version is detected until you install it.
  37. v0.10.3
    • Changed Tables in Slate read like tables. Cells render bold, links, and code the way the rest of the document does. Wide tables scroll sideways inside their block instead of squishing every column, and a tall cell only stretches its own row.
    • Changed Right-click recents. Open, reveal in Finder, or move to trash from the recents list. Saved slates all live in one place now.
    • Changed Spoken homographs sound right. Words like "read", "lead", and "wind" get the right pronunciation for the meaning instead of guessing.
    • Fixed Cmd-W closes the front window. The shortcut had quietly stopped working for non-Pro accounts; now it works for everyone, on any window.
  38. v0.10.2
    • Added A bigger, calmer Slate. Each block in the canvas now shows who wrote it: you, the assistant, or a pending suggestion you can accept, reject, or take over. A small pill lets you undo a single edit without rolling back the whole reply. Streaming feels less like a wall of text and more like a document being written.
    • Added Token usage in the footer. See how much of your conversation is fresh, cached, or response, and which model actually answered. Reset context with a button or by saying so out loud when you want to start a new train of thought.
    • Added Zoom Slate the way you'd expect. Pinch on a trackpad or use Cmd-=, Cmd--, Cmd-0. Diagrams and equations rescale with the prose. Your zoom level sticks across sessions.
    • Added Frontmatter, files, and drag-drop. A collapsible metadata strip above the document round-trips YAML cleanly. The assistant can read your other open files when you ask. Drop a markdown file onto Slate or double-click one in Finder to open it.
    • Changed Markdown that finally looks like markdown. Tables read as proper tables. Horizontal rules are visible. Diagrams stay readable instead of getting clipped. Mousing over a diagram or equation no longer freezes page scrolling.
    • Changed Auto-save is opt-in. Slate stays an ephemeral scratchpad by default. Turn auto-save on in Settings when you want changes to persist without thinking about it.
  39. v0.10.1
    • Changed Better recording on AirPods and Bluetooth headsets. A new audio capture path makes dictation reliable on Bluetooth, with no more dropped starts or missing first words when you press to talk.
    • Added Other apps duck instead of muting. While you're dictating, music and video drop to a comfortable background level instead of being silenced. Adjust how far they duck in Settings → Dictation.
    • Added Permissions pane. Microphone, Accessibility, and Automation permissions now have their own home in Settings, with a cheerful animation when you grant one and a tap-to-replay if you want to see it again.
    • Added Replay any paragraph in Slate. Click any paragraph of a spoken assistant reply to hear just that bit again, with a moving highlight that follows along.
    • Changed Spoken paths and emails come out right. Saying "slash users slash mike" or dictating an email address now produces a clean path or email instead of literal words.
    • Changed Polish. Hotkeys section in General collapses to a single row that opens the full Hotkeys sheet. Subtle haptic snaps on key toggles. Menu bar popover dismisses cleanly when you click another menu bar item.
  40. v0.10.0
    • Added Slate. A live, voice-driven canvas where your assistant's replies stream in as proper markdown: headings, lists, code, tool calls. Save when something's worth keeping; ephemeral by default. Pro feature, in active development. Learn more →
    • Changed A more alive Dashboard. Every pane now shows what your settings actually do: Augments lists each active directive, AI Writing rotates concrete before/after examples, Voice Output gets a clearer "Speaking" indicator, Dictionary has a try-it sandbox, and Assistant has a live trigger-word preview.
    • Changed Bigger, clearer keycaps. Hotkey symbols across the app render at proper size and shape, easier to read at a glance, and the same visual everywhere they show.
  41. v0.9.8
    • Changed A clearer Dashboard. The settings window is now split into a Dashboard (where you live: AI Writing, Dictionary, Assistant, Voice Output, Characters, Personas, Augments) and a Settings overlay for the configuration bits (General, Dictation, HUD, Models, History, License). The sidebar regroups around what you actually do: Writing, Assistant, Voice Stack. No more one-item sections.
    • Changed The Dashboard always opens at launch. Removed the toggle, so opening Voice Mode always brings up the Dashboard so you land somewhere instead of having to hunt the menu bar.
    • Changed AI Writing toggles can't thrash anymore. Rapidly turning AI Writing or AI Correction on and off used to queue several seconds of model reload per flip. The toggles now coalesce to a single load or unload, and disable while the model is loading so you can't pile work on top of work.
    • Changed "Enter" command handles trailing punctuation. Saying "send the message enter" or "send the message. Enter." both work, and the period stays attached to the right word in the typed text.
    • Changed Polish across the Dashboard. A bit more space under the Super Voice title and between row labels and their descriptions; AI Writing's sub-options stay visible (dimmed) when AI Writing is off so you can see what's there.
  42. v0.9.7
    • Changed Workspace opens quietly after onboarding. Finishing onboarding still surfaces the Workspace so first-run users land somewhere useful, but it no longer steals focus from whatever you switch to next. It's there waiting in the Dock when you want it.
    • Changed HUD marked experimental. The recording visualizer is flagged as experimental in the Workspace while we tune the styles and full-screen modes.
  43. v0.9.3
    • Changed A redesigned Workspace. The settings window has a fresh look: a cleaner sidebar with light and dark modes, a redesigned About and License experience, a refined HUD with a new extra-small size, and a Characters space alongside Personas. Active augments and personas appear in a strip across panes so you always know what's on.
    • Added A guided first run. New users are welcomed with a short tour of the assistant and sensible defaults, so Voice Mode is ready to use the moment you open it.
  44. v0.9.2
    • Changed AI Correction is now free. The on-device LLM cleanup pass on dictation, plus highlight-and-correct on selected text, are now part of the free tier, no subscription required. Pro continues to cover the voice assistant, character voices, personas, and augments.
  45. v0.9.1
    • Fixed Voice output reads file paths and code references naturally. Backtick-wrapped names like notes/meeting.md are now spoken as "notes forward slash meeting dot md" instead of leaking raw symbols. Em-dashes also read as natural pauses.
  46. v0.9.0
    • Added Voice Mode integrates with AI coding assistants. Your assistant can now read your Voice Mode settings, switch personas, swap voices, and adjust augments, all with your permission. Connect from Settings, and revoke access any time.
    • Added Ask the assistant about Voice Mode. A new Voice Mode Help augment teaches the assistant how Voice Mode works. Turn it on and ask things like "how do I change my voice" or "what's a persona". Activate from Workspace → Augments.
    • Added More assistant options. Additional backends are now selectable in Workspace → Assistant.
    • Fixed Settings are safer. Voice Mode now keeps a backup of your settings and recovers automatically if the file ever gets corrupted, so you don't lose your configuration.
    • Fixed Switching voices no longer interrupts speech. Changing the voice backend used to leave you mute while the new one downloaded. The previous voice now keeps working until the new one is ready.
    • Fixed Voice downloads show progress. Larger voice models now display real download progress and surface errors in the Models pane, instead of silently churning.
  47. v0.8.5
    • Changed Polish across personas and AI Correction. Small refinements to behavior and defaults.
  48. v0.8.4
    • Changed Edit & Correct works without a selection. Press the edit hotkey with nothing highlighted and your spoken instruction becomes a command. Say "translate the following to German: hello how are you" and the result pastes at the cursor. With text selected, the rewrite-in-place behavior is unchanged.
    • Changed Edit & Correct is the default in the menu bar. The tray popover now surfaces Edit & Correct instead of AI Correction. AI Correction is still available in Settings but is labeled experimental.
    • Fixed AI Correction is much faster. Correction lands in about a second and a half now, instead of occasionally timing out and skipping the pass.
    • Fixed AI Correction is harder to trick. Phrases like "translate to German" or "ignore previous instructions" inside your dictation are now treated as words to transcribe, not commands to run.
  49. v0.8.3
    • Added Say your character's name to wake the assistant. When a character or persona is active, say their name to activate the assistant: "Iris, what time is it" or "hey Iris, look this up". The default wake word still works too.
    • Added Character name in the UI. The popover, drawer, and top-notch bar now show the active character's icon and name.
  50. v0.8.2
    • Added Menu-bar update indicator. A small dot appears on the mic icon when a newer version is available.
    • Changed Multilingual dictation, for real. The default speech-to-text engine now supports all 99 advertised languages. Chinese, Japanese, Korean, Arabic, Hindi, Thai, Vietnamese, Turkish, and Persian all transcribe correctly.
    • Changed Smart formatting respects your language. Spoken-number and spoken-operator rules no longer corrupt non-English dictation.
    • Fixed Downloads resume where they left off. Interrupted multi-gigabyte downloads no longer restart from the beginning.
  51. v0.8.1
    • Added Edit & Correct (Pro). Highlight text in any app, hold the edit hotkey, speak an instruction, and the selection is rewritten in place. Works across native Mac apps and most Electron editors.
  52. v0.8.0
    • Changed Improved local AI model. Correction and the local-AI assistant both move to a new model with meaningfully better punctuation, grammar, and list formatting.
    • Fixed Live download progress when activating an AI model. The percentage now advances as the download runs instead of sitting at "Loading…".
  53. v0.7.0 Pro subscriptions
    • Added Pro subscriptions are live. Monthly and Annual plans. Every paid plan starts with a 14-day free trial, cancel anytime.
    • Added License activation in-app. Paste your SVM-… key in the License pane to unlock Pro. Up to 3 Macs per license.
    • Added 14-day offline grace period. Travel, fly, work without wifi, and your Pro features keep working.
  54. v0.6.0 Async assistant
    • Added Keep dictating while the AI thinks. Dictation works immediately after speaking, even while a request is still processing.
    • Added Task panel. See running, completed, and failed assistant requests with live timers. Cancel or replay any of them.
    • Added Top-notch mode. The assistant lives in the MacBook Pro notch area. Works on non-notch Macs too.
    • Added Voice cancel. Say "cancel that", "never mind", or "stop" to interrupt running tasks.
  55. v0.3.0 Fully native
    • Changed One native app, no extra setup. Dictation and AI are now a single Mac app: no background helpers, no dependencies to install, no setup after dragging it into Applications.
    • Changed Faster, more efficient speech-to-text. Transcription is noticeably quicker and easier on the battery.