Starts when your call does
As a call rings or begins, a small prompt offers to record. One click.
Every call transcribed on your Mac, with every speaker named. No bot joins. Nothing gets uploaded.
Free, with everything. No account.
Requires macOS 26 or later on Apple silicon. iPhone and iPad coming soon.
No bot in your call
No upload to the cloud
No account, no subscription
No notes to take
TalkKeep records what your Mac hears: your mic and the meeting’s sound. No extra guest, no invite, no plug-in.
Every voice is told apart on your Mac, then named: straight from Zoom, Slack, Meet, or Tencent Meeting, or once by you and remembered in every meeting after. It learns your own voice, too.
Switch between English and Chinese mid-sentence. TalkKeep keeps both, with real punctuation, in readable paragraphs.
Search every meeting at once, or type @ to keep to one person or one app, then click a line to play it from exactly there.
Claude Code, Codex, and other AI agents on your Mac can search every meeting and quote it word for word, with a link to the moment. Read-only, and off until you turn it on.
As a call rings or begins, a small prompt offers to record. One click.
A minute before a meeting, it offers to join and record. The recording takes the event’s name.
Live captions stay above every window while you record.
When the call hangs up, or a minute goes quiet, it offers to stop.
In the app or in the browser.
Speaker echo is removed from your mic.
What you say on mute is left out, in every app.
Saved every minute, even through a crash.
Add names and terms once, and transcripts write them the way you do.
Meetings listed by day, with a calendar to jump to any date.
Titled by the app and the people in it.
Markdown, text, subtitles, or the audio.
Recording, transcription, and speaker recognition all run on your device.
Record the meeting in the room on iPhone or iPad, transcribed right on the device. Every meeting, from any of them, is on all of them.
Nothing appears on their side, because TalkKeep doesn’t join the call. Recording laws vary, so tell people when you record; TalkKeep reminds you before your first recording.
No. TalkKeep only goes online to download its speech models once, and to sync through end-to-end encrypted iCloud if you turn that on.
Thirty languages and 22 Chinese dialects, even when a sentence switches between them.
Yes, all of it, on the Mac, iPhone, and iPad: there is no account to make, no subscription, and nothing held back.
Any call your Mac plays: Zoom, Google Meet, Microsoft Teams, Slack huddles, FaceTime, Webex, Discord, Tencent Meeting, Feishu, DingTalk, WeChat, and meetings in a browser tab. The website version also reads names and your mute state from Zoom, Slack, Google Meet, and Tencent Meeting.
Voices are told apart on your Mac and recognized again in later meetings, including your own. The website version also reads who is speaking in Zoom, Slack, Google Meet, or Tencent Meeting, so names come from the call itself; name anyone once and that name wins everywhere.
Most of them send a bot into the call and your audio to a server. TalkKeep does neither: it records what your Mac already hears, transcribes it on the device, and keeps every meeting in a library only you can open. Nothing to upload, no account to make, nothing for others in the call to see.
Only if you let them. Turn on command-line access in Settings → Developer, and Claude Code, Codex, and other agents on your Mac can search and read your transcripts, but not change them or play the audio. An agent may send what it reads to its servers, so this stays off until you turn it on.
Same app. The website version can also read names and mute state from your meeting app, which the App Store doesn’t allow.
A Mac with Apple silicon and macOS 26 or later, or iOS and iPadOS 26 or later. The speech models are a one-time 2.3 GB download.
Free, with everything. No account.
Requires macOS 26 or later on Apple silicon. iPhone and iPad coming soon.