Changelog
What changed in each MaxTalk version. This is a developer preview; check for and install new versions in Settings › About.
0.2.4 · 2026-10-27 · Translate key, rewrite what you just said
- Translate key: set one in Settings › Shortcuts, hold it and speak in any language, and it types in the language you chose, such as speaking Chinese and typing English. Fillers go, the meaning is translated the way a native speaker would write it, the tone follows the app you're in, and names, code and numbers stay as they are.
- Rewrite what you just said: right after something is typed, hold the AI key without selecting anything and say “make that more formal” or “translate the last sentence into English”, then click Replace. If the text can't be found in the field, nothing is changed and the new text is copied instead.
- Open at login: turn it on in Settings › General; MaxTalk then appears in System Settings › General › Login Items.
- Feedback: Settings › About › Feedback exports a diagnostic package to your Desktop, so you can look inside before sending it. It has the version, system and permission states, which models you use, and the log and crash reports with personal details taken out; no API keys, nothing you said and no recordings.
- Capsule drawer: the drawer is as big as its cards, with no empty space when there are only a few.
- Several displays: the voice capsule (HUD) opens on the display the pointer is on and remembers where you drag it, on an external display too. A position you dragged it to before is reset once with this update.
- Faster: with the decision model on in Labs, an action no longer waits for the text cleanup, about half a second sooner.
- New address: the website is now maxtalk.top, and updates come from there; the old address keeps working.
0.2.3 · 2026-10-27 · Capsule drawer
- Capsule drawer: press ⌃⌥Space and your ideas and what's waiting for your OK slide out from the screen edge; press again, press Esc or click elsewhere to put them away, and the keyboard goes back to the app you were in. While it's open, hold the AI key to add an idea. Drag a card into any text field; file ideas away or turn them into to-dos, and add what's waiting with one click.
- Settings: Settings › Shortcuts › Capsule drawer changes or turns off its key and picks the edge it slides out from (left, right, top or bottom). If ⌃⌥Space is already one of your keys, the drawer key starts off and you can pick another.
- Several displays: the drawer opens on the display the pointer is on.
0.2.2 · 2026-10-27 · Notarized by Apple
- Notarized by Apple: MaxTalk is now signed with a Developer ID and notarized by Apple, so it opens with a double-click after you download it, with no trip to System Settings to click “Open Anyway”. Because the signature changed, grant Microphone, Accessibility and Input Monitoring once more; macOS may ask once whether MaxTalk may read the keychain (choose Always Allow).
- Fixed: quitting MaxTalk while the speech model was loaded was recorded by macOS as a crash.
- Crash reports: Settings › About shows the crash reports on this Mac and opens them in Finder. They stay on your Mac, are never uploaded, and have API keys and your user name taken out.
- Permissions: when you click Grant and macOS shows no prompt (you said no before, or an older version's entry is still there), MaxTalk opens the right page of System Settings and tells you to remove MaxTalk from the list and add it again.
- Updates: when a version can't be installed from the app (its identifier changed, for example), MaxTalk says to download it from the website and offers a button that opens the download page.
0.2.1 · 2026-10-25 · Final app identifier, snippets
- Final app identifier: MaxTalk's identifier is now com.maxtalk.desktop. Settings, history, speech models and API keys move over the first time it opens (macOS may ask once whether MaxTalk may read the keychain; choose Always Allow); system permissions need granting again. Coming from 0.2.0, install this one by hand.
- Snippets: write “trigger = text” in Settings › Speech; say just the trigger and the text is typed as written, without AI.
- Dictation: cleaned up for the kind of app (casual in chat, written in mail and documents, names and English kept as said in code); with text typed as heard, “new line” and “new paragraph” work too.
- Conversations: on a calendar or to-do card, ask “what about Wednesday?”; delete a whole conversation from the Conversations page.
- Updates: a download can be cancelled; if MaxTalk won't open after an update, you can go back to the previous version.
- Faster and cheaper: the fixed part of each request can be cached by the provider (now Claude too); Settings › Privacy › Usage shows the cache hit rate.
0.2.0 · 2026-10-25 · Conversations, calendar lookups and updates
- Updates: check, download, install and reopen from Settings › About; every download's signature is checked first; optional daily checks (off by default). Coming from 0.1.0, install this one by hand.
- Conversations: the main window has Inbox, Conversations and History. A question and its follow-ups are one conversation you can go on with by typing or speaking; “what can you do?” is answered from what MaxTalk really does.
- Clearer results: calendar and to-do lookups get their own card with a timeline and free time; opening, searching or setting the volume says what was done; to-dos, reminders and events can be undone right after adding; opened apps and lookups are one line each under History › Activity.
- Notes: “note in Notes: …” writes to Apple Notes, or by default to Obsidian or Bear (Settings › Actions).
- Email: say a name and the address is found in Contacts; drafts open in Mail, or in Gmail or Outlook on the web.
- Calendar and to-dos: “am I free tomorrow afternoon?”, “what's left on my to-do list today?”, summed up on your Mac and never sent to AI.
- Media and Mac: play, pause, next track, volume and mute, dark mode, screenshots, lock screen.
- Fixes: live text while you speak (Labs) stopped partway every time, because of a flaw in the Whisper library MaxTalk uses, now worked around; Jev's instructions now say that words meant for someone else, commit messages and product ideas are not actions; no flash of the last card when the talk key goes down; no made-up song credits in live text; the HUD stays readable over dark windows; “from
” only where it means something.
0.1.0 · 2026-10-24 · First preview
- Two keys: the AI key (Right ⌥ by default) works out what you mean and does it; the dictation key (Right ⌘ by default) types what you say at the cursor. Record any combination for either key, choose hold-to-talk or press-to-start for each, and double-tap in hold mode for hands-free.
- Typed the moment you let go: also in WeChat, Chrome, Claude and other apps. Dictation offers AI cleanup (the words as heard if it takes over a second) or text exactly as heard. If it didn't land, press ⌃⌘V to type it again.
- On-device speech recognition: Whisper runs on your Mac and works offline. The recommended model, Large v3 Turbo (quantized), about 574 MB, downloads in one click during setup. Custom vocabulary supported.
- AI: rewrite selected text, answers that stream in with follow-ups, and ideas. Bring your own API key: OpenAI, Claude, Gemini, DeepSeek, Qwen, Kimi, GLM, OpenRouter, or local Ollama and LM Studio.
- Actions: reminders, to-dos and events (added after you confirm, and undoable), email drafts; opening pages, searching 20 sites, opening apps and System Settings panes, running Shortcuts; message drafts for WeChat, Slack and more, never sent for you.
- Labs (off by default): live text while you speak; the Jev decision model for faster intent; finding the contact for message drafts.
- Known limits: Apple silicon and macOS 13 or later only; not yet notarized by Apple; no automatic updates yet.