◐ Shell
reader mode source ↗
Trace
Back to Trace

Press

Press kit

Boilerplate, logos, and contact for writers, reviewers, and podcast hosts covering Trace.

Last updated
15 June 2026
Contact
[email protected]

At a glance

  • What it is: a native macOS menu-bar app that records meetings and transcribes them on-device.
  • Hero claim: meeting transcripts that never leave your Mac. No meeting bots, no accounts, no cloud.
  • Three things that set it apart: it's local (on-device, nothing leaves your Mac), it's fast (Apple-silicon-optimised and on-device, so no upload and no queue), and it has key moments (flag decisions mid-call with one keystroke, inline in the transcript).
  • How you use it: one keystroke to start, one to flag a key moment, one to stop. Out comes a markdown transcript.
  • Price: £9.99 one-time on the Mac App Store. No subscription.
  • Availability: macOS 14+ on Apple silicon.

Short boilerplate

Trace is a native macOS menu-bar app for meeting capture and transcription. It records mic and system audio with one keystroke, transcribes on-device using a local speech model, identifies each speaker on the call so every line carries a numbered label (Speaker 1, Speaker 2, …), and lets you flag key moments live with ⌘⇧K so they land inline in the markdown transcript at the exact timestamp they happened. The resulting transcript is plain markdown ready to paste into Notion, Obsidian, or any AI notetaker. Trace itself doesn't do AI. It just hands you the raw, contextual text. Everything stays on your Mac: no meeting bots, no accounts, no telemetry. The only required network request is a one-time download of the speech and speaker models on first run. An optional calendar connection is available for users who want sessions named after the meeting they're in: Trace can read the Mac's calendar entirely on-device (covering any account), or connect to Google directly. It is opt-in, read-only, and off by default.

Long boilerplate

Trace is a Mac menu-bar app for on-device meeting transcription with three design pillars: everything is local, transcription is fast, and the user can flag key moments live. It records the microphone and any audio playing through the Mac (Zoom, Teams, Google Meet, Slack, a phone on speaker) as two separate tracks, transcribes them locally using an on-device speech model, and writes a clean markdown transcript with speaker labels and timestamps. Mic-track lines carry your editable label (Microphone by default); system-audio lines run through an on-device speaker identifier so each distinct voice on the call gets its own numbered label (Speaker 1, Speaker 2, and so on). All processing stays on the Mac.

The interaction that sets Trace apart is the key moment: hit ⌘⇧K mid-call, type a short note, hit Enter. The flag lands inline in the transcript as a gold blockquote at the exact timestamp you pressed it. You stay in the conversation, and the record you hand off afterwards has explicit, human-authored signposts for the moments that actually mattered. Key moments travel with the transcript into whatever AI notetaker you paste it into (ChatGPT, Claude, the notetaker of your choice), so the AI's summary is anchored to what the human in the room said was important. Trace doesn't summarise for you. It just makes sure the record is accurate and easy to hand off.

Recordings stay on-device. The only required network request is a one-time download of the speech model plus two smaller models used for speaker identification (about 500 MB combined) from Hugging Face on first run. After that, recording, transcription, and speaker identification never touch the network: no servers, no accounts, no telemetry. Sessions live inside Trace's sandboxed container under ~/Library/Containers/, browseable in Finder and deletable any time. You can also pause mid-recording with ⌘⇧P, rename past sessions from the menu bar, and enable Stealth Mode to hide the floating pill and silence every sound the app plays. Trace runs on macOS 14 and later on Apple silicon, ships as a sandboxed app on the Mac App Store, and costs £9.99 as a one-time purchase with no subscription.

An optional calendar connection is available for users who want recordings to auto-name after the meeting they're in. There are two sources, chosen in Preferences. Mac Calendar reads upcoming events on-device through Apple's EventKit, covering every account already set up in the Calendar app (iCloud, Google, Exchange, and others) with no sign-in and no network request at all. Google Calendar connects one account directly using the read-only calendar scope and never writes back. With either source, a small pill appears one minute before each meeting offering to start recording. The calendar connection is off by default and changeable at any time from Preferences; Google is the only source that sends a network request after the model download, while Mac Calendar stays entirely on-device. No other integrations, and no plans for any that would route audio or transcripts off the Mac.

Features, one-liners

The three pillars:

  • Yours, locally: plain audio and markdown files on your disk. No servers, no accounts, no telemetry. Transcription runs entirely on-device.
  • Fast and local: a local speech model transcribes on your Mac's Apple silicon. No upload, no queue.
  • Flag key moments live: hit ⌘⇧K mid-meeting to mark a decision with an optional note. The flag lands inline in the transcript at the exact timestamp.

And the supporting interactions:

  • Out of the way until you need it: one keystroke summons Trace over any app; another dismisses it.
  • Markdown you can paste anywhere: Notion, Obsidian, ChatGPT, Claude, any AI notetaker, or plain text.
  • Every transcript, one click away: the menu bar keeps a quiet list of your sessions, each one one click from copy or reveal-in-Finder.
  • Speaker-aware: each distinct voice on the call gets its own numbered label (Speaker 1, Speaker 2, …) in the order they first speak. Speaker identification runs on-device alongside transcription.
  • Meeting-aware (optional): connect your calendar (Mac Calendar on-device, or Google) and Trace names sessions after the meeting you're in and nudges you one minute before each one starts. Read-only, opt-in, off by default.

Facts for fact-checking

  • Speech models: two on-device options, a faster model and a more accurate model (the default) that downloads a larger weight set the first time you use it.
  • Speaker identification: Pyannote-based segmentation model plus a WeSpeaker v2 embedding model, both running on-device through the same FluidAudio pipeline.
  • Model runtime: the FluidAudio Swift package, running on-device (no Python, no subprocess).
  • Audio capture: AVAudioEngine for mic, Core Audio process taps for system audio. Tracks are written as 16 kHz mono float32 WAV files.
  • Hotkeys: Carbon RegisterEventHotKey (Mac App Store safe, does not require Accessibility permission), user-remappable.
  • Default shortcuts: ⌘⇧R toggle recording, ⌘⇧K flag a key moment, ⌘⇧P pause or resume, ⌘⇧H show or hide the floating pill, ⌘⇧? reveal the recap of the last few minutes.
  • Distribution: Mac App Store (sandboxed, hardened runtime).
  • Dependencies: Swift and Apple frameworks, plus FluidAudio for the on-device speech runtime. The only required network traffic is a one-time model weight download from Hugging Face on first run.
  • Optional calendar sources: Mac Calendar via EventKit (on-device, no network, covers any account in the Calendar app) or Google Calendar (read-only, OAuth with PKCE). Off by default. Used for session auto-naming and a one-minute pre-meeting pill nudge. Changeable from Preferences.

Downloads

Contact

[email protected]