- Saved voiceprints are no longer wiped when you remove a language pack or the speaker model is incomplete.
- Choosing Re-transcribe while an import is transcribing now waits its turn instead of running alongside the import.
Changelog
What changed in each release. New versions install right from the app.
v0.15.6
- Speech recognition is faster: dictated text and live meeting captions appear sooner, with identical results.
v0.15.5
- After you release the key, dictated text appears faster and more consistently — no more occasional multi-second waits.
v0.15.4
- Dictation no longer ends a sentence at every pause; punctuation now follows the meaning, so one sentence stays one sentence.
- AI polish no longer adds paragraph breaks of its own.
v0.15.3
- Dictation now ignores nearby voices and background noise, keeping only what you say.
- AI polish now writes spoken numbers as digits, cases product names and acronyms properly, and tidies punctuation.
- Text appears sooner after you release the key.
- The dictation pill is now just a waveform that moves with your voice, and looks the same whether you hold or tap.
- An occasional crash at the moment a dictation finished.
v0.15.2
- Dictation now frees its scratch memory when it finishes, so keeping the model resident costs much less.
- The dictation pill is leaner: no helper text, sized to its content, and the polishing star now twinkles.
- The dictation pill sometimes failed to appear, or vanished right after appearing.
- Releasing the key while the speech model was still loading was ignored, leaving dictation stuck.
- The dictation shortcut occasionally needed two presses to respond.
v0.15.1
- Release notes in the update prompt now show English first, then Chinese.
v0.15.0
Dictation is here: hold the right ⌘ key in any text field, speak, release, and the text is typed for you. Names it learned from your meetings are recognized while you dictate.
- Dictation: hold the right ⌘ key to speak and release to paste; tap it once to lock, tap again to finish, so a ten-minute dictation never needs a held key.
- AI polish cleans up fillers and self-corrections from your dictation, in three strengths; pick its model separately under AI Provider.
- AI Provider now lets you assign one model to meeting summaries and another to dictation, instead of a single default.
- Launch silently: start without the main window and stay in the menu bar.
- Microphone, system audio recording and accessibility permissions are now managed together under Privacy & Data, each labeled with what uses it.
v0.14.0
This release does one thing: transcription memory drops from over ten gigabytes to under one, and it is released the moment transcription finishes.
- Transcribing an hour-long meeting now peaks around 1 GB instead of over ten, and drops back to about 100 MB when it finishes.
- Transcribing several meetings in a row no longer keeps piling up memory.
- The Chinese model pack has a new version that is smaller to download and lighter to run, with the same recognition quality; update it under Settings › Models.
v0.13.1
- Meeting summaries are generated differently: each fact appears once, the headline is the outcome, action items only include things someone actually took on, and suggestions are no longer written up as decisions.
- Summaries of English meetings are now written natively in English rather than reading like a translation.
- The playback speed menu no longer shows a gray box around it, and the speed button no longer has a dark rim.
- Checked action items now show a clear checkmark instead of a solid block.
- The current-word highlight during playback no longer extends past the word.
- Scrolling the summary no longer leaves several lines highlighted at once.
v0.13.0
This release is about knowing who spoke and trusting what the notes say: better speaker recognition, and every number in a summary can be traced back.
- Live captions now show speaker names as people talk — anyone in your voiceprint library is labeled the moment they speak.
- Right after you stop recording, the live captions are readable as a draft; the full transcript replaces it automatically when ready.
- Microphone-only recordings of in-person meetings now separate speakers and recognize people you know.
- Lines whose speaker is uncertain get a "doubtful" badge — one click reassigns or confirms them.
- Summaries now list entities not found in the transcript and numbers that only appear in low-confidence lines, with a Locate button to jump to them.
- Names the summary couldn't match can be added to the dictionary in one click, so future transcripts spell them right.
- Live captions are steadier and faster: fewer retracted words, earlier final lines, no backlog even in long meetings.
- Transcripts get cut off mid-sentence far less often, and swallowed second halves are recovered automatically.
- Summaries of English meetings are now reliably written in English instead of occasionally coming out in Chinese.
- The Add Term panel in Settings was redesigned with clearer fields; press Return to submit.
- Clickable rows in Settings now respond when you click on their text.
- Long explanatory text in Settings no longer gets cut off.
- Summaries no longer fail entirely when the background line is missing.
v0.12.1
- Anthropic couldn't connect before — it now talks to the official Messages API directly.
- Summaries now work through API relay gateways, and Test Connection no longer reports false positives.
- A batch of AI Provider settings fixes: long error messages and API keys now display properly, editing no longer switches vendors by accident, and the dialog title stays put as the sheet grows.
v0.12.0
- Added anonymous usage statistics — feature usage only, never any meeting content or personal information; turn it off anytime in Settings → Privacy & Data.
- One-click MCP install now supports Codex.
- Settings got a full redesign — clearer sections, and model providers you can tell apart at a glance.
- Redesigned the waiting screens for transcription and summary generation — it’s clearer what’s happening at each step.
- Summaries for European-language meetings are now written in the language actually spoken.
v0.11.1
- Due to a system permission identifier change, macOS will re-request screen recording and microphone access after this update — MeetKeep walks you through it once; your data and license are untouched.
v0.11.0
- The free trial is now 7 days — and it starts with your first meeting, not the moment you install.
- Meeting-app detection is more reliable — no more occasional stale results.
- Clearer MCP errors for agents — unknown tools now list what is available.
v0.10.5
- Export a diagnostics bundle from Settings → Help & Feedback when something goes wrong — logs only, never your recordings or meeting content.
- If MeetKeep didn't shut down cleanly, it offers to export diagnostics on the next launch.
v0.10.4
- Fixed a handful of small interface issues.
v0.10.3
MeetKeep tells people apart more accurately, and the Voiceprints page has been rebuilt.
- A detail page for each person: see the clips their voice is built from, play them back, and remove any one of them.
- When identifying an imported recording, people MeetKeep already knows are named automatically.
- Speaker separation is noticeably more accurate, and twice as fast.
- Read-aloud enrollment redesigned — the text highlights as you speak, and you can stop any time.
v0.10.2
- Redesigned Library — meetings, recordings and imports can be viewed separately.
- You can now record the audio of a single app.
- Files can be dropped anywhere on the window to import.
- Imports appear in the list right away, and failed ones can be retried on the spot.
v0.10.1
- Speaker separation is always on — the toggle is gone from Settings.
- Fixed microphone permission being impossible to grant on 0.10.0. Just grant it again after updating.
v0.10.0
- MeetKeep notices when a meeting was recorded in the wrong language and offers to redo it in the right one.
- Live captions now work for English and European languages.
- Editing transcripts, opening meetings and search are all noticeably faster.
- Memory is released when idle, so MeetKeep sits lighter in the background.
- Fixed playback being silent right after a recording finished.
v0.9.5
Urgent fix: 0.9.3 / 0.9.4 crash on launch. The old app never runs, so auto-update cannot reach it — please download and install manually.
- Fixed the crash on launch that affected both 0.9.3 and 0.9.4.
- Microphone + meeting app now follows whichever client is actually detected.
v0.9.4
Chinese recognition is much more accurate. You will need to download the Chinese model pack again (about 3.5GB); the old one can be removed under Settings › Models.
- Personal lexicon: fix a name once and MeetKeep remembers it.
- Muting in MeetKeep mutes your meeting app too.
- Mixed Chinese-English, and terms like SKU or API, now come out right.
- Fewer duplicated and imagined lines in two-track meetings, and faster transcription.
- Fixed re-transcribing an imported meeting always failing.
- Fixed a crash when double-clicking a transcript line to edit it.
v0.9.3
- Fixed an issue with license activation.
v0.9.2
v0.9.0
The first public beta. Meetings record themselves, Chinese and English both transcribe on your Mac, and every line of the notes clicks back to the audio — none of it leaves the machine.
- Recording starts when you join and stops when everyone leaves. Waiting rooms never trigger it.
- Chinese and English transcribe on your Mac — click any word to play from there.
- It tells who said which line, and once you enroll a colleague’s voice their name shows up in every meeting after that.
- AI notes write outcomes only — what was decided, who owns it, when it is due, what is still open — and every line clicks back to the exact words.
- A one-page visual summary you can copy as a single image.
- Three ways to record — mic only, mic plus one app, mic plus everything — and you can import existing recordings.
- Search titles, transcripts and notes together, then click a result to land on that moment.
- Claude Code and Codex can read your meetings: read-only, off by default.