Support

Getting started

Your first recording

The first-launch onboarding walks you through granting microphone and system audio recording access. Once you’ve made it through that:

1. Press ⌘⇧R, or click the Trace icon in the menu bar and choose Start Recording.
2. A small pill appears above the Dock. Talk, join a call, play a video: everything audible gets recorded.
3. Press ⌘⇧R again to stop. The pill switches to a progress ring while Trace transcribes on-device. The first ever transcription downloads the speech model it needs. After that everything runs offline.
4. When the ring completes, click the pill’s name field, type a name, and hit Enter. Click Copy to put the transcript (markdown with timestamps, speaker labels, and any key-moment notes) on your clipboard, then paste it anywhere. If you named the recording while it was running, Trace shows a small “is ready” notice instead of the pill; click it to open the recording. You can change what happens at this point in Settings → Feedback → When a transcript is ready.

The menu bar vs the pill

• Menu bar icon (top right of your screen): the quick way in. Start a recording, jump to a recent session, open Settings. Available even when you’re not recording.
• The Trace window: where your recordings live. It opens on a Today screen with your upcoming meetings, what you recorded recently, and a box to ask a question of everything. See Finding your way around below.
• Floating pill (above the Dock, only visible during a session): the live state. Shows your elapsed time, a waveform, and buttons for key moments, pause, and stop.

You can hide the pill mid-session with ⌘⇧H or turn it off entirely in Settings → Feedback. Hiding it does not affect the recording.

Moving the pill

The pill sits above the Dock by default. If it covers something you need to see, drag it. Press anywhere on it, including its buttons, and move it where you want. A drag never triggers the button you started on, so you can pick the pill up by Stop without stopping the recording.

While you drag, dashed guides show where the pill normally sits, and it snaps back onto them if you drop it close. Trace remembers the spot for each display, so reconnecting a monitor puts the pill back where you left it on that screen. If the display is gone, the pill returns to its usual place on the screen you’re using.

Settings → Feedback shows where the pill will appear and has a Reset to put it back.

Permissions

Trace needs two macOS permissions, both of which first-launch onboarding walks you through:

• Microphone captures your voice.
• System Audio Recording captures audio from your other apps (Zoom, Teams, and so on, so the other people on your call).

Trace does not record your screen for either of these. The one feature that touches your screen is the optional Screenshots feature, which grabs a region you choose, only when you ask, and uses macOS’s Screen Recording permission (it asks the first time you capture). See Screenshots below.

To re-run the onboarding flow (useful for double-checking both grants in one place), open the system menu bar while Trace is active and choose Window → Welcome to Trace.

To change a grant directly, open System Settings → Privacy & Security, find Trace under either category, and toggle. Changes take effect after the next Trace launch.

Recording

Starting and stopping

• Global shortcut: ⌘⇧R toggles.
• Menu bar icon: click → Start Recording or Stop Recording.
• System menu bar (when the Dock icon is visible): File → Start Recording / Stop Recording.

By default, stopping via keyboard requires a double press within a couple of seconds. This guard prevents a stray ⌘⇧R mid-meeting from killing your recording. You can turn it off in Settings → Recording. Menu clicks stop immediately either way, since a click is always deliberate.

Auto-start

When another app takes your microphone, which usually means a call is starting, Trace offers to record it. A small prompt appears with Start recording and No, and a chevron that remembers a choice for that app: Always start or Never ask. An app set to Always start begins recording on its own, with the usual start sound and pill, so you never forget to hit record.

This is on by default. Turn it off, or change a saved per-app choice, in Settings → Meetings → Auto-start. The prompt (and any Always apps) pause while stealth mode is on.

Recording just the meeting app

By default Trace records all of your Mac’s system audio. When you’re on a call it can narrow the system track to just your meeting app, so music or a video playing in another tab stays out of the transcript. Trace picks the meeting app automatically when a call starts, and if a different app starts a meeting mid-recording it asks before switching what it captures. You can change which app it follows from the pill during a recording, or set a default in Settings → Recording. When the meeting app quits, Trace widens back to all system audio on its own.

The consent reminder

If you like to announce that you’re recording, Settings → Meetings has a Remind me to announce recording toggle (off by default). With it on, a quiet reminder appears as recording starts; press Consent noted and Trace stamps the time you announced it into the session’s metadata. It never blocks the recording and disappears on its own after a few seconds. It’s a convenience, not legal advice.

Key moments

A key moment is a timestamp you flag during a recording, optionally with a short note. Trace inlines each one into the transcript as a blockquote. Use it to mark “this is the important part” for whoever reads the transcript later, whether that is you, a colleague, or a summariser. When you use Trace’s own on-device summary, your key moments are always carried through, so they are the parts it is told to keep.

• Global shortcut: ⌘⇧K.
• The pill shows a note field right after each key moment. Type a short note, hit Enter. Skip it and Trace just records the timestamp.
• The auto-dismiss delay is configurable in Settings → Recording. Default is 10 seconds, with 5 / 20 / 30 / 60 second options, plus Never (the prompt stays up until you press Enter or explicitly dismiss it).
• Every keystroke resets the countdown, so the delay is really “how long after you stop typing the prompt stays up” rather than a hard cap on how long you have to write.
• You can turn the note prompt off entirely in Settings → Recording if you only want the timestamps, without a note.

Screenshots

While recording, you can grab a region of your screen and attach it to the transcript at that moment. It is for the things audio alone misses, like a slide, a chart, or a diagram someone is presenting.

• Global shortcut: ⌘⇧S. It is also in the pill’s overflow menu (the ⋯ button) as Capture the screen, with Capture a new area (⌥⌘⇧S) to pick a different region.
• The first capture in a recording asks you to draw the area to grab, then Trace reuses it for the rest of that recording, so the next captures are instant. You can set the area ahead of time in Settings → Screenshots.
• Each screenshot sits inline in the transcript where you took it, and once transcription finishes Trace gives it a short title and summary so you can scan what it showed.
• Screenshots need macOS’s Screen Recording permission. It is the only feature that uses it, audio recording does not, and macOS asks for it the first time you capture. You may need to quit and reopen Trace after allowing it. You can grant it any time in Settings → Screenshots.
• The images are saved with the session on your Mac, like everything else Trace records.

The live recap

While you’re recording, Trace can show a running transcript of what’s being said in a drawer above the pill. Press ⌘⇧? to open and close it.

Treat it as a preview rather than the finished article. It runs its own streaming speech model so it can keep up with the conversation in real time, while your actual transcript is produced afterwards by whichever model you picked in Settings → Models. The final transcript is the accurate one.

• Scroll up and it holds still. Reading back through what was said no longer pulls you down to the latest line. A button in the drawer’s header counts what has arrived while you were reading, and clicking it catches you up.
• You can select and copy a line. Highlight text in the recap and press ⌘C.
• It follows the meeting’s language. The recap understands 40 languages, including French, German, Spanish, Japanese and Mandarin, and switches to the meeting’s language automatically. It can also translate the meeting into your own language as it happens; see Translation below.
• The first time you record, Trace downloads the recap’s model (about 640 MB) and caches it. After that it works offline like everything else.

Pausing

⌘⇧P pauses and resumes. Paused time is excluded from both the elapsed timer and the transcript.

Key moments are ignored while paused (there’s no audio at that instant to point at); the pause button icon reminds you.

Erasing something that shouldn’t be recorded

If something is said that shouldn’t be on the record, press ⌘⇧E. Trace silences the last ten seconds of both audio tracks and takes them out of the transcript.

The pill shows how much is about to go and puts a cancel button where the stop button normally sits, wrapped in a countdown ring. Press ⌘⇧E again inside that countdown to reach another ten seconds further back, and again to keep going. Nothing is erased until the countdown ends, so the cancel is your way out. Once it ends, the audio cannot be recovered.

The stretch it erases is measured from the moment you first press, not from when the countdown finishes. However long you take deciding how far back to go, anything said while you were deciding is kept.

• The erased seconds become silence. The recording keeps its original length, so timestamps, key moments and playback all still work.
• It works while paused, unlike key moments.
• It is also in the pill’s overflow menu (the ⋯ button) as Erase the last 10 seconds.
• How far each press reaches is in Settings → Recording, under Erasing. The shortcut itself is in Settings → Shortcuts.
• In stealth mode it still works but shows nothing, which also means there is no way to cancel it.
• If you have used Transcribe so far, its partial file is rebuilt from the audio as it now stands.

The meeting-ended prompt

When the app using your microphone ends its call, which usually means the meeting is over, Trace asks whether to stop the recording. It never stops on its own, so you choose Stop or keep recording, and the prompt appears within a few seconds even when the pill is hidden. Stealth mode holds it back, along with everything else Trace would otherwise put on screen. You can turn it off in Settings → Meetings.

Discarding a recording

Not every recording is worth keeping. Discard throws one away instead of transcribing it.

You can discard a recording from the meeting-ended prompt, from the pill’s menu while it is recording or transcribing, and from its row in the Trace window. Discarding drops any transcription already queued and skips the naming, summary and Shortcut that would otherwise follow.

If the Undo button on the pill is not pressed within five seconds the recording is moved to the Trash, so you can still recover it from there until you empty it.

Switching to the app that took the microphone

If Trace is recording all system audio and another app takes the microphone mid-recording, the pill offers to follow it: “<App> took the microphone. Switch?”. Switch scopes the recording to that app. Dismissing the offer with the ✕ leaves the recording as it is and also stops that app counting towards the meeting-ended prompt for the rest of the recording, so dictating in another app no longer ends with Trace asking whether your meeting has finished. Never turns the offer off for that app for good; you can undo that in Settings → Meetings → System audio.

Naming sessions

When a recording starts, a small name field opens on the pill. Type a name and hit Enter and it is saved straight away, so if the app is interrupted mid-meeting the recording keeps your name instead of a timestamp. You can turn this prompt off in Settings → Recording if you’d rather start without it.

You can rename during a recording by clicking the pill’s timer, and rename any past session by clicking its title in the Trace window or from the menu bar list (hover the row, click the chevron, choose Rename). When a recording finishes, the completion pill also offers a name field. With AI features on and automatic naming off, clicking that field offers three suggested names; pick one with the mouse or with ⌘1, ⌘2 or ⌘3, and it goes into the field as ordinary text you can still edit. The same three suggestions appear when you rename a recording later from its page. If you have a calendar connected, ongoing calendar events auto-name your sessions.

Back-to-back recording

You don’t have to wait for a transcription to finish before starting the next meeting. Stop a recording and Trace hands it to a background queue, freeing the app straight away, so you can start the next call while the previous one is still transcribing. Up to two recordings can be in flight at once.

When you record over a transcription, the pill splits: the transcribing job pinches off into a small token on the right with a queue count, and the new recording takes the centre. The overlapped sessions are saved and copied as they finish, and you can name them later from the menu bar.

Importing existing recordings

You can transcribe a file you already have, not just what Trace records live. Drag an audio or video file onto the Trace window, or click Import audio at the bottom of the sidebar, and Trace reads through it and produces the same transcript and summary it would for one of its own recordings. Large files are handled in the background with a progress ring on the recording’s row, so the app stays responsive while it works. Drop several at once and they queue: the row that is transcribing shows its progress and the others show that they are waiting.

Transcripts

Where they live

By default, sessions live inside Trace’s own storage:

~/Library/Containers/info.traceapp.trace/Data/Library/Application Support/Trace/sessions/

You can put them somewhere else. See Choosing where sessions are kept below.

Each session is a folder named with an ISO timestamp and an optional slug (e.g. 2026-04-22T10-15-33-team-standup). Inside:

• transcript.md: markdown with timestamps, speaker labels, and any key-moment notes. This is what the Copy button puts on your clipboard. Its layout is yours to change (see The Markdown template below).
• transcript.json: the structured source the markdown was rendered from.
• summary.md: an on-device summary of the meeting. Only written if you ask for one (see On-device summaries below).
• meta.json: the session’s identity. The name you gave it, its start and end time, its length, and a log of each transcription attempt. It’s written even if transcription never runs.
• mic.wav / system.wav: the audio tracks (kept or removed according to your audio-retention setting; see Audio retention and cleanup below). If you have turned on compression, these are mic.m4a and system.m4a instead.

The easiest way to open a specific session is from the menu bar: hover the row, click the chevron, and choose Open (opens transcript.md in your default markdown handler), Reveal in Finder, Copy, Rename, or Delete.

The system menu bar also has File → Open Sessions Folder and File → Open Recent for the same flow.

Finding your way around

The Trace window opens on Today. The top of the page lists the rest of today’s calendar meetings, with a Start recording button on the one about to begin, then what you have recorded recently grouped by day. A recording that is still running has its own controls on its row, so you can stop, pause, mute or mark a key moment without finding the pill. The sidebar keeps the five recordings you opened most recently under Recent, and All recordings is the whole library with sort and filter chips.

Back and forward work like a browser’s, including between a recording’s Transcript, Summary and Ask tabs: ⌘[ and ⌘], or the arrows at the top left. ⌘P opens a command palette that jumps to any recording or action by typing, and ⌘K opens Ask. All four can be rebound in Settings → Shortcuts.

Folders

Drag a recording onto a folder in the sidebar to file it. A recording is in one folder or none, and anything unfiled stays under Unfiled at the bottom of the sidebar, so you can see what is still to sort. Click + next to Folders to make one; a folder’s page has its own list, an overview paragraph you can ask Trace to write (with AI features on), and an Ask box that reads only that folder’s recordings. Folders are a way of grouping; tags are still there for anything that belongs in more than one place.

Choosing where sessions are kept

Settings → Privacy & Storage → Session library shows the folder Trace writes into. Pick anywhere you like: Documents, an external disk, or a folder your sync tool already watches.

When you pick a new folder, Trace offers to move recordings from the old folder to the new one. It copies each session, checks it arrived intact, and only then removes the original. A folder of the same name already at the destination is skipped and reported rather than merged or replaced. If you decline the move, those sessions stay where they are and still show in your history.

A few things worth knowing:

• If the folder isn’t there, recording still works. An external disk might be asleep, or a cloud folder’s helper might not have started. Trace keeps the session in its own storage and moves it across when the folder is back.
• If Trace loses access to the folder, choose it again. Deleting the folder, or replacing it with a new folder of the same name, ends the permission macOS gave Trace, and no amount of waiting brings it back. Recording carries on into Trace’s own storage, and picking the folder again in Settings hands the permission back.
• Trace doesn’t sync anything. If you point two Macs at the same synced folder, Trace won’t merge or lock files. It makes no promises about both machines writing at once, so be careful when using Trace simultaneously across multiple machines.
• Cloud folders are worth testing first. Trace writes to a recording continuously, and a service like Google Drive’s File Stream may not handle that. Make a short recording into the folder and play it back before trusting a long one.
• You can’t move the library mid-call. Finish the recording or transcription that’s running first.

The downloaded speech and summary models, and the cross-session voice registry, stay in Trace’s own storage.

Search, find, and edit

• Search: the search box on All recordings ranks results by how well they match, with the line that matched shown under each row, so the right meeting is usually at the top. A match on a recording’s name outranks a passing mention. With AI features on, search also matches on meaning, so a topic in your own words finds the meeting even if nobody used those words; that downloads a second, smaller model the first time it runs. With AI off you get the keyword ranking, which needs nothing extra.
• Command palette (⌘P): from anywhere in the Trace window, ⌘P opens a palette that jumps to a recording by name and runs actions (start or stop a recording, flag a key moment, summarise the last meeting, open Settings) from the keyboard.
• Find in a transcript (⌘F): with a transcript open, ⌘F opens a find bar. Return jumps to the next match, Shift-Return the previous, Escape closes it.
• Edit a line: double-click any transcript line to fix what was misheard; a single click plays the audio from that point. Edits update both the transcript and its Markdown, and the session is marked as hand-edited, so a later re-transcribe warns you before replacing your changes.
• Speakers panel: the bar above the transcript shows each person’s share of the talking. Open it for talk time, the longest uninterrupted stretch, and everything you can do to a speaker: rename, merge, or hear a snippet of their voice while the audio is still kept.

Translation

Trace can translate meetings into your language, entirely on your Mac, using Apple’s on-device translation models. It needs macOS 15 or later; on macOS 14 the translation controls stay hidden.

• During a meeting. When the live recap detects a language that isn’t yours, a translate button appears on the pill. Turn it on and each line of the recap is translated into your language the moment its sentence completes, with the original kept underneath if you want it. The button’s card holds the master switch and the per-line options.
• After a meeting. A Translate control sits beside Transcript, Summarise and Ask on any finished session. Pick a language and Trace produces a translated copy alongside the original, keeping every speaker and timestamp, so click-to-play and renames keep working. A banner names the language pair, and Show original flips back. Copy and export follow whichever version you’re reading. The original transcript is never touched.
• Your language is set during onboarding and lives in Settings → Translation. It is what transcripts translate into, and what the live recap compares the meeting’s language against.
• A meeting you translated live automatically gets a translated finished transcript, and a Settings option extends that to any foreign-language recording.
• The first translation to or from a language downloads that language’s pack through macOS’s own consent sheet; after that it works offline.

Tags

Give a recording as many tags as you like from its page in the library: click Add tag and Trace autocompletes against every tag you’ve used before. Filter All recordings by tags from the filter chips (a session must carry every tag you tick). Rename, merge, or delete a tag everywhere in Settings → Tags.

If AI features are enabled, Suggest tags for recordings in Settings → AI (off by default) reads each new recording and proposes up to five tags. It names the things that came up (a company, a product, a standard, a place) rather than summing up the meeting, so a tag stands a chance of fitting the next recording too, and a proposal that carries a tag you already have folds onto that tag instead of adding a near-duplicate beside it. The proposals sit in the tag row as dashed chips: click one to accept it, or its x to turn it down for good. A proposal for a tag you don’t have yet is marked NEW, and accepting it adds that tag to your list. Nothing is applied without you unless you also turn on Apply suggested tags automatically, which sits under the first toggle. Tags you added by hand are never touched.

Ask questions about a meeting

With AI features enabled, you can ask Trace a plain-language question in three places, and where you ask sets what it reads:

• Today has an ask box that looks through every recording you have, for when you can’t remember which meeting something was in.
• A folder’s page has an ask box that reads only that folder.
• A recording’s Ask tab, next to Transcript and Summary, reads only that recording.

Trace answers entirely on your Mac, using the same local model as summaries. It works through your library the way a person would, looking things up and opening what it needs before it answers, rather than answering from a single search. Each answer carries citation chips. Press one to play the recording from that moment, and press it again to show the line in the transcript. Conversations are kept, so a question about one recording is saved with it, and questions about everything or a folder sit under Chats in the sidebar for you to pick up later. Summaries carry the same chips, with one after each point, so you can validate each claim or jump back for context.

Trace can also brief you before a call. If you have a calendar connected and AI features on, an event starting in the next 15 minutes gets a Pre-call brief banner at the top of the Trace window. Trace searches your whole library by meaning as well as by words, opening the recordings that look relevant, and writing up what matters from them.

Spotlight

Turn on Index transcripts in Spotlight in Settings → Privacy & Storage (off by default) and finished recordings show up in macOS Spotlight search, matched by title, transcript text, and named speakers. The index is macOS’s own local Spotlight store on your Mac. Turning the toggle off removes Trace’s entries again.

Speaker labels

• The mic track uses one label, which represents you. Default Microphone; editable in Settings → Speakers.
• The system-audio track runs through Trace’s on-device speaker identifier. Each distinct voice on the call starts with a numbered label (Speaker 1, Speaker 2, and so on, in the order they first speak). You can put real names to those voices after the call, covered next.
• In-person meetings get speakers too. When a recording’s system track has no speech (everyone was in the room, so every voice arrived through the microphone), Trace runs the speaker identifier over the mic track instead and separates the voices there. A solo dictation still comes out under your own single label.
• When two tracks overlap (you speak while someone else is speaking on the call), the mic track wins the timestamp, so your line comes through verbatim.

Naming and merging speakers

When a recording finishes, the completion pill shows a row for each voice Trace heard. Type a name on each row and Trace rewrites the transcript with the names in place. A play button on each row plays a short sample so you can tell who is who.

• Merge two rows when the identifier split one person in two, and separate them again if you merge by mistake.
• Edits save as you go, the moment you click out of a field, so there is no extra save step.
• The numbered source (transcript.json) is never overwritten, so naming is always reversible. Re-transcribing a session starts the names over.
• Reopen this any time from a past session in the menu bar (chevron → Organise speakers).

Trace also remembers voices across calls. Once you have named someone, the next recording that hears them suggests that name on the unnamed row, and you confirm with a tap. Nothing is ever filled in for you. Manage everyone Trace has learned in Settings → Speakers, where you can rename, merge or forget a voice. Only names and a numeric voiceprint are stored, never any audio, and only on your Mac.

Naming a recording

By default the pill leaves your keyboard alone. Click its name field when you want to type, then press Enter to save the name. Trace used to take the keyboard the moment the field opened, which could catch a sentence you were writing in another app and turn it into the name of your meeting.

If you liked naming a recording the instant it finished, turn on Allow the pill to take keyboard focus when input is required in Settings → Feedback → Keyboard focus.

The key-moment note field ignores that setting and always takes the keyboard. It only opens when you press ⌘⇧K, so you can type the note straight away.

Copying to clipboard

Trace never copies anything to your clipboard on its own. On the “ready” pill, clicking Copy copies the full transcript.md. Paste it into Notion, Obsidian, ChatGPT, Claude, Slack, or anywhere else.

You can copy any past transcript from the menu bar: chevron → Copy transcript.

If you would rather Trace copied automatically, Settings → Feedback → When a recording finishes has a toggle for the transcript and one for the summary. Both are off, because a recording that finishes while you are working in another app would otherwise replace whatever you had just copied.

On-device summaries

The completion pill shows a Summarise button (in the slot the old Reveal button used). Press it and Trace summarises the meeting on your Mac and writes a summary.md next to the transcript. The button then reads Open summary. Your clipboard is left alone unless you switch the copy on in Settings → Feedback → When a recording finishes.

• It is manual by default. Nothing is summarised until you press the button, though Settings → AI has a Summarise automatically after transcribing toggle if you’d rather it just happened.
• Opening a session’s Summary tab never starts a run. If there is no summary yet, the tab tells you which model would run and how big the download is, and waits for you to press the button.
• It runs entirely on-device, on Trace’s own local language model, so the audio and the summary never leave your Mac.
• The first summary downloads that model once (a few GB) and caches it; after that it works offline.
• The key moments you starred during the call are always carried into the summary, under their own heading.
• You can summarise a past session too, from its row in the menu bar (Summarise, then Open summary or Copy summary).
• The instructions Trace gives the model live in Settings → AI, where you can save a named template for each kind of meeting (a stand-up, a client call, a one-to-one) and switch between them. The built-in templates are editable too, and a reset puts any of them back. An Action items as Markdown checkboxes toggle (off by default) writes the summary’s action items as real checkboxes that apps like Obsidian treat as tasks.
• In a session’s Summary tab the text is selectable, so you can highlight and copy exactly what you need, and each section has its own copy button for grabbing just the action items or a single part.
• Summaries, suggested names, tags and Ask replies follow the language of the transcript. A meeting held in German gets a German summary; there is nothing to configure. If you’ve translated a transcript (see Translation above), these features read the translated copy instead, so their output comes out in your language.

Using your own AI model (custom endpoint)

If you’d rather run the AI features through a model you control, point Trace at any OpenAI-compatible server in Settings → AI: switch the summary model to Custom endpoint, then enter the server’s base URL, an optional API key (stored in your Keychain), and the model name. A Fetch models button lists what the server offers, and Test connection confirms it works end to end.

• Works with a local server such as LM Studio or Ollama running on your Mac, or a hosted API you have an account with.
• Once set, summaries, naming, tags and Ask all run through that endpoint. The transcript text is sent to the server you chose, so pick one you trust; with a local server nothing leaves your machine.
• Live transcription always stays on-device, whichever endpoint you use.
• Reasoning models that think before they answer are handled, and long summaries are never cut short.
• If the server returns an error, Trace shows its message, and the error is included when you copy diagnostics.

Fixing mis-heard names and jargon

Speech models mishear unusual names, acronyms and product names. Rather than fix the same word every time, add a rule to the word-replacement dictionary in Settings → Dictionary: list the variants you see (for example clawed, Clawde, clod) and the spelling you want (Claude).

• Rules apply automatically to every new transcript, in the transcript.md, the transcript.json and the copy on your clipboard.
• Matching is whole-word and case-insensitive, so a rule never changes text in the middle of another word.
• Re-transcribing an older session picks up the dictionary too. It does not rewrite transcripts you have already made.

Pausing a transcription

Transcribing a long recording works your Mac hard. If you need that power back, press the pause button on the left edge of the transcribing pill (or the Pause button on the progress view in the History window) and Trace stops transcribing straight away.

• The paused state shows in the menu bar, with a Resume button to pick it back up whenever you’re ready.
• Recording is unaffected; you can pause a transcription and start a new recording at the same time.
• The interrupted track starts transcribing again from the beginning when you resume.

Hiding a transcription in progress

Transcribing a long meeting can take minutes, and the progress pill sits on screen the whole time. Press the hide button on the transcribing pill (the eye with a line through it, just right of pause) and the pill goes away while the transcription carries on. The completion pill appears as normal when it finishes.

Hiding applies to that one transcription, not to the pill in general. The next recording in the queue shows its pill as usual, and retrying a session you had hidden brings its pill back.

Retrying failed transcriptions

The audio is always written to disk before transcription runs. If transcription fails (out of memory, model cache corrupted, etc.), the session shows up in the menu bar with a Retry Transcription action. Click it and Trace re-transcribes from the saved audio. No need to record anything again. A failed session keeps its audio as uncompressed WAV and will not compress until a transcription succeeds, regardless of your auto-compression settings.

Audio retention and cleanup

The audio is a 16 kHz float32 WAV that uses substantial disk space, so by default Trace moves mic.wav and system.wav to the Trash once transcription succeeds, where you can still recover them until you empty it.

Change that in Settings → Privacy & Storage with the Keep audio setting:

• Until I delete it: keep the raw audio for good.
• Delete after transcription: move it to the Trash as soon as the transcript is ready (the default).
• For 1 day, For 1 week, or For 30 days: keep it for a window, then clean it up.

When you pick a timed window, two more controls appear:

• When sessions expire, remove chooses the scope: Audio only (the default; transcripts and summaries always survive) or Entire recording (the whole session, transcript and all, goes to the Trash once it ages out).
• Automatic cleanup is the switch for the timed sweep, and it never runs until you turn it on. Review and turn on… first shows exactly which recordings would be removed right now, so nothing disappears unseen. Changing the window or the scope pauses the sweep again until you re-approve it.

Everything Trace cleans up goes to the Trash rather than being deleted outright, and a failed transcription always keeps its audio so you can retry.

Compressing your recordings

If you keep your recordings, they add up fast. Trace can compress your recordings to keep disk usage down.

A compressed recording plays back the same to most human ears, but because it is a lossy version of the original using the compressed audio to re-transcribe a recording may result in a worse quality transcript than the original.

There are two options to control how Trace compresses your recordings.

• Compress now… compresses everything already on disk. Settings → Privacy & Storage shows how much disk space your recordings use and how many are still uncompressed.
• Compress audio sets a standing rule for new recordings: Never (the default), After transcription, After 1 day, After 1 week, or After 30 days. It only offers windows shorter than your Keep audio window, since a recording that has already been deleted cannot be compressed.

If you ask Trace to transcribe a compressed recording again, it warns you first.

Compressed tracks are mic.m4a and system.m4a in the session folder, playable in QuickTime, Music, or anything else on your Mac.

Exporting and sharing

The Export sheet

Open a finished session and click the export button to export the transcript, or the summary once one exists, as PDF, Word (.docx), Web page (.html), Subtitles (.srt or .vtt), or Spreadsheet (.csv). Subtitles and the spreadsheet come from the transcript only, since they need per-line timings. PDF has two extra options: a Confidential or Privileged watermark, and an optional password that encrypts the file. Everything is generated on your Mac.

The Markdown template

The layout of transcript.md is editable in Settings → Export: a template built from tokens like {{title}}, {{date}}, {{summary}}, {{transcript}}, {{key_moments}} and {{action_items}}. It shapes the file on disk, and with it everything downstream: the clipboard copy, the dragged-out file, the folder copy, and what gets sent to other apps. (The Export sheet’s PDF, Word and HTML formats lay themselves out and don’t use the template.)

The same pane holds three related toggles, all off by default: Add YAML frontmatter, Link key moment timestamps to the audio (each key-moment timestamp becomes a link that reopens the session in Trace at that second), and Speaker names as Obsidian links (named speakers become [[wikilinks]]).

Drag a session out

Drag the tick on the completion pill straight into Finder or another app. You get a Markdown copy of the transcript named after the session.

Save every transcript to a folder

Settings → Automation has Save a copy of each transcript and summary to a folder (off by default). Choose a folder and every finished transcript, and its summary once written, is copied there with a filename pattern like {date} {title}. Point it at an Obsidian vault or a synced folder and your notes app picks everything up on its own.

This writes copies with a readable filename, and it doesn’t include the audio. To put the whole session, audio included, into your own folder, see Choosing where sessions are kept above.

Send to another app

Also in Settings → Automation: Send finished recordings to an app (off by default). Pick Obsidian, Bear, Drafts, or Things, and whether to send the transcript, the summary, or both. A send button then appears on finished recordings in the History window and the menu bar. Obsidian receives the full text every time; the others fall back to putting very long transcripts on the clipboard with a note.

Export a key moment as a clip

If a session still has its audio, every key moment in the transcript offers Export clip: a small WAV of the seconds around the moment (5 before and after by default, up to 30 each way), with mic and call audio mixed. Handy for sharing the exact quote. The button only appears while the audio is still on disk, so check your retention setting if it’s missing.

Redact before sharing

The redact button on a finished session scans the transcript on your Mac for names, email addresses, phone numbers, postal addresses, ID numbers and dates of birth, then shows you every match so you choose what to keep or drop. It writes a masked copy, transcript.redacted.md, next to the original, and can mask the summary too. The original transcript is never modified.

Keyboard shortcuts

All global shortcuts are configurable in Settings → Shortcuts, and any of them can be cleared to turn its hotkey off entirely (the feature still works from the menu bar). The defaults are:

Inside the Trace window there are a few more, and the first four can be rebound in Settings → Shortcuts too: ⌘K opens Ask, ⌘P opens the command palette, ⌘[ and ⌘] go back and forward, and ⌘F finds inside an open transcript.

⌘⇧R is claimed for as long as Trace is running. The rest are claimed only while a recording is in progress, so every other app keeps those combinations while Trace is idle. The screen capture (⌘⇧S) and erase (⌘⇧E) shortcuts follow the same rule, and the recap drawer chord stays claimed while the completion pill is up.

If a shortcut doesn’t fire, another app (a system one, or a launcher like Raycast) has probably claimed the same combination. Open Settings and pick a different combo. Trace uses the Mac App Store safe Carbon API for global shortcuts, so it does not need Accessibility permission.

Apple Shortcuts and automation

Beyond the global hotkeys above, Trace plugs into Apple’s Shortcuts app and Spotlight.

Actions in Shortcuts and Spotlight

Search for Trace in the Shortcuts app or Spotlight and you’ll find actions for Start, Stop, Pause, Resume, Mute and Unmute the mic, Capture a key moment (with an optional note), Get last transcript, Get last summary, and Summarise last meeting. The control actions behave exactly like the matching hotkeys, and the read actions hand over the file, so they are safe for a transcript of any length.

Run a Shortcut when a transcription finishes

In Settings → Automation, turn on Run a Shortcut automatically when I finish a transcription and type the exact name of a macOS Shortcut. When a transcription completes, Trace puts the transcript markdown on the clipboard and runs that Shortcut with the transcript as its input, so a “save to Obsidian” or “post to Slack” Shortcut needs no file handling of its own. A Run now button fires it against your most recent transcript so you can test without recording. Off by default.

The same pane holds Save to a folder and Send to an app (see Exporting and sharing above), and the command-line tool’s install line.

If your Shortcut sends the transcript somewhere, that copy leaves your Mac. Everything Trace itself does stays on-device.

Settings

Open with ⌘, from anywhere inside the app, or from the menu bar → Settings…. Settings is a sidebar of panes in five groups:

App

• General: appearance (match your Mac, or pin light or dark), the Dock icon, the quit confirmation (Cmd-Q checks before Trace closes; on by default), and factory reset.
• Feedback: when the pill shows (during recording and for key moments), where the pill is positioned (with a Reset back to the default spot), whether the pill may take keyboard focus, what happens when a transcript is ready (ask for a name every time, ask only if it needs one, a quiet notice, open it in the library, or nothing), the two clipboard copies (both off by default), per-effect sound toggles, hints, and milestone notifications.
• Shortcuts: rebind or clear every global shortcut, and the window shortcuts for Ask, the command palette, back and forward. The system menu bar mirrors whatever you pick.

(The version bought direct from our website adds a Licence pane here too.)

Capture

• Recording: input device with a live level meter, the system-audio source, echo cancellation, double-press to stop, the name-on-start prompt, and key-moment options.
• Meetings: auto-start, the meeting-ended prompt, the offer to follow an app that takes the microphone and the apps you have answered Never to, the consent reminder, and calendar sources.
• Speakers: your own track’s label, and the people Trace remembers across calls (rename, merge, forget).
• Tags: every tag you’ve used, with rename, merge, and delete-everywhere.
• Dictionary: the word-replacement rules for names and jargon.

Content

• Models: the transcription model picker, downloaded models and their disk use, keep-in-memory, and the advanced download bypass.
• AI: the master Enable AI features switch, the summary model and its options (including a Custom endpoint for running the AI features through your own server), automatic naming and tagging, and your summary templates.

Output

• Export: the Markdown template and tokens, YAML frontmatter, timestamp links, and Obsidian wikilinks.
• Automation: run a Shortcut on finish, save to a folder, send to an app, and the command-line tool.

Privacy

• Privacy & Storage: the session library folder, offline mode, stealth mode, Spotlight indexing, audio retention and automatic cleanup, voice-clip retention, and diagnostics, plus the privacy policy and terms links.

A few panes worth a note:

Echo cancellation

Cancel system audio from the microphone (in Settings → Recording) removes audio that plays out of your speakers and leaks back into your microphone, so the system audio doesn’t end up in the transcript twice. It’s off by default, because it adds a little time after each recording while the audio is processed, and with headphones you likely don’t need it. If you record without headphones on your Mac’s speakers, turn it on. After you stop, the pill shows a short Removing echo phase while it finishes the job. The setting is read when a recording starts, so changing it takes effect on the next one.

Stealth mode

Hides the pill and silences all sounds. The menu-bar icon also stays static (it does not reveal recording state), and the auto-start prompt pauses. Use when screen-sharing, presenting, or in any context where you’d rather not broadcast that you’re recording. It lives in Settings → Privacy & Storage.

Offline mode

One switch in Settings → Privacy & Storage that stops Trace making any network request it would start on its own, like calendar syncs and model downloads. While it’s on, a small pill shows next to the Settings button in the History window so you don’t forget. It’s Trace choosing not to reach out, not a firewall: it can’t stop other apps or macOS itself.

Privacy and permissions

What Trace records

• The mic track (your voice).
• The system-audio track (anything audible on your Mac while a recording is active).

What leaves your device

Nothing, by default. Audio, transcripts, and metadata never touch the network. The only network requests Trace ever makes are the one-time model downloads. The live recap fetches its own streaming model (about 600 MB) the first time you record, the speech model arrives the first time you transcribe, and the local summary model the first time you use an AI feature. After that, recording, transcription, summaries, and questions all run entirely offline.

The one exception is one you set up yourself. If you point the AI features at a custom endpoint (see “Using your own AI model” above), the transcript text they work on is sent to the server you chose. Audio never leaves your Mac either way, and live transcription always runs on-device.

If you turn on the optional calendar connection, Trace reads the event titles and timings of calendars you explicitly allow. That’s it. It never writes to your calendar and never reads descriptions, attendees, locations, or conferencing URLs. With the Mac Calendar source this all happens on-device; with Google it uses a read-only connection. See the Privacy Policy for the full breakdown.

Consent and recording laws

Recording and consent laws vary from country to country and region to region. It is your responsibility to obtain any consent that is required and to follow the law wherever you are.

Trace records and transcribes everything locally and never uploads your audio or transcripts. That keeps your recordings private, but it does not change your obligation to get consent where it is needed. If you want a nudge, the consent reminder in Settings → Meetings shows a small prompt each time recording starts (see The consent reminder above).

Calendar integration (optional)

Off by default. Pick a source in Settings → Meetings. When a source is connected, Trace:

1. Auto-names sessions that overlap with a calendar event.
2. Shows a small pill one minute before a meeting, offering to start recording. No auto-start: you click, or you don’t.

There are two sources:

• Mac Calendar reads your events on-device through macOS, covering every account already in the Calendar app (iCloud, Google, Exchange, and others). Nothing is connected and nothing goes online. Trace asks for read-only calendar access the first time you pick it; revoke it any time in System Settings → Privacy & Security → Calendars.
• Google Calendar connects one Google account directly. Disconnect at any time from Settings → Meetings. The stored tokens are wiped from your Mac immediately. You can also revoke access from Google’s side at myaccount.google.com/connections; Trace will notice on the next refresh and quietly return to its no-integration state.

Uninstalling

Drag Trace.app from /Applications to the Trash. Your recordings stay inside Trace’s container at ~/Library/Containers/info.traceapp.trace/. Delete that folder too if you want them gone for good. If you moved the session library somewhere else (see Choosing where sessions are kept), your recordings are in that folder instead and deleting the container won’t touch them.

Troubleshooting

Trace isn’t hearing anything

Check three things in order:

1. System Settings → Privacy & Security → Microphone: Trace must be enabled.
2. System Settings → Privacy & Security → System Audio Recording: Trace must be enabled to capture audio from your other apps (Zoom, Teams, browser, and so on).
3. System Settings → Sound → Input: your actual input device is correct. If you have AirPods or a USB mic, macOS sometimes switches inputs mid-session.

My system audio is not being recorded

The first time you record, macOS asks for System Audio Recording permission. If you declined it, your microphone is still captured but audio from your other apps (Zoom, Teams, and so on) is not, so the system track stays silent.

To fix it, open System Settings → Privacy & Security → System Audio Recording, turn Trace on, then start a new recording. Trace can’t re-ask once you’ve declined, so this has to be done in System Settings.

The other person on my call is recorded twice, or my microphone track sounds echoey

This is what Cancel system audio from the microphone fixes. When you record on speakers, the audio coming out of your Mac leaks back into the microphone and gets recorded a second time, so the same words end up on both tracks and your microphone track sounds echoey.

The toggle is off by default. Turn it on in Settings → Recording and start a fresh recording, since the setting is read at the start of each session. The short Removing echo phase after you stop is it doing the work.

The global shortcut isn’t working

Another app (system utility, launcher like Raycast, or another menu-bar tool) has probably claimed the same combination. Open Settings → Shortcuts inside Trace and pick a different combo. No restart needed; the system menu bar updates to match the new chord immediately.

If it is one of the recording-only shortcuts, check that a recording is actually running. Trace releases those combinations while it is idle, so they do nothing until you start recording.

Transcription is slow

Trace runs an Apple-silicon-optimised speech model on your Mac’s neural engine, so transcription happens locally with no upload and no queue. Longer meetings naturally take longer to transcribe. Trace prepares the model in the background at launch, so a finished recording no longer has to wait for the old warm-up step.

If something seems off, the session will surface a Retry Transcription option in the menu bar. Clicking it re-runs transcription from the raw audio without needing to re-record.

Why is the pill saying “Downloading model” or “Optimizing for your Mac”?

The first time you transcribe on a new Mac, Trace has to get the speech model ready, and that is two separate waits. The pill tells you which one you are looking at.

Downloading model is the first. Trace transcribes on your Mac, so the model has to be on your Mac, and it is a large file: about 950 MB for the Accurate model, 650 MB for Balanced, 220 MB for Fast. The pill shows the size and a percentage so you can see it moving. On a slow connection this is the longest part of the whole thing. It happens once per model.

Optimizing for your Mac is the second. Once the file is down, macOS compiles the model for your specific chip so it can run on the Neural Engine. That can take a couple of minutes on a recent Apple Silicon Mac, and longer on older or memory-constrained machines. Trace usually does this quietly in the background after you first choose a model, so you will often not see it at all. After that, transcriptions reuse the compiled cache and go straight to “Transcribing”. You will only see it again if you upgrade macOS, change Macs, or delete the cached model in Settings → Models.

You can start a new recording while either one is running. Nothing about recording is blocked by it. Hit your record shortcut (⌘⇧R by default) or use the menu bar, record your meeting, and Trace transcribes it as soon as the model is ready. Your audio is safe on disk in the meantime.

The model download won’t complete

The first transcription downloads the speech model. Behind a corporate proxy or security software that inspects TLS, that download can fail partway. Settings → Models → Advanced has a Bypass secure download option: turn it on and Trace fetches the model over a connection it controls, relaxing the checks a TLS-inspecting proxy can break. It’s off by default and flagged as reducing security, so only use it if the normal download won’t go through. Once a model is on disk it always loads locally and never re-checks the server.

If the download stops with a message about Offline mode, that switch is on in Settings → Privacy & Storage; turn it off and retry.

How to make the transcription more accurate

Trace ships with two transcription models. The default, Accurate, gives you a literal, verbatim transcript and does the best job on accents, jargon, and quiet audio. It runs entirely on-device and is the most faithful record of what was actually said.

If you’d rather have your transcript faster, switch to the Fast model in Settings → Models. It trades raw word-for-word precision for speed, and its output is designed to be handed to a language model afterwards, which reads between the lines and recovers the meaning even when individual words come out slightly off. Pasted into ChatGPT, Claude, or any other LLM with a prompt like “summarise this meeting”, the result is usually as accurate as a human transcript, often more so. The choice applies to every new transcription, including retries on past sessions.

A few things that help accuracy on either model:

• A decent mic. AirPods and built-in laptop mics work, but a wired headset or a USB mic noticeably improves what the model hears.
• Turn on echo cancellation if you record on speakers. The Cancel system audio from the microphone toggle in Settings → Recording stops the audio playing out of your speakers from being recorded back onto your microphone track. It’s off by default, since it adds a little processing time after each recording and isn’t usually needed with headphones.
• Heavy crosstalk degrades both tracks. The on-device speaker identifier handles normal overlap fine. We know you can’t always control how a meeting flows, but it’s worth knowing that several people talking over each other is the hardest thing for any speech model to untangle.
• Domain jargon and proper nouns are the next hardest. If the meeting is full of acronyms or unusual names, add them to the word-replacement dictionary (see Fixing mis-heard names and jargon) so Trace corrects them automatically on every new transcript, and fix any one-offs by hand. For a transcript you already have, double-click any line to correct it in place.

Which languages does Trace support?

More than English. Both transcription models are multilingual, and both detect the spoken language on their own, so there is nothing to switch on before you record. The live recap follows along too; it understands 40 languages and switches to the meeting’s language automatically.

The default Accurate model is the one to use for any language other than English. It can recognise around 99 languages and picks the right one automatically. We tested it on German, French, Spanish and Italian, and it handled all four accurately.

The Fast model covers 25 European languages, among them German, French, Spanish, Italian, Dutch, Polish and Portuguese. It detects the language for you as well, so it is a good choice for those languages when you want speed. For a language outside that set, or when you are unsure, stay on Accurate, which reaches much wider.

Accuracy still depends on the audio, the same as it does in English. Clear speech, a decent mic and limited crosstalk all help, and a widely spoken language with a neutral accent comes out cleaner than a rarer one or a strong regional accent.

If a recording comes out wrong, or you would like us to check a language we have not named here, get in touch (see Contact below).

The AI features follow along too. Summaries, suggested names, tags and Ask replies come out in the language of the transcript, so a German meeting gets a German summary without any setup.

And if you’d rather read a meeting in your own language, Trace can translate it, live during the call or after it has finished (see Translation above).

Does Trace work without internet?

Yes, once the models are cached. The first recording downloads the live recap’s streaming model and the first transcription downloads the speech model, both one-time. After that, recording and transcription are fully offline.

If you’ve connected Google Calendar, that source specifically needs the network. The Mac Calendar source reads on-device and works fully offline. To keep Trace completely offline, set the calendar source to Off (or use Mac Calendar) in Settings → Meetings.

There’s also a dedicated Offline mode switch in Settings → Privacy & Storage: while it’s on, Trace won’t start any network request of its own, and a small pill by the Settings button reminds you it’s active.

A session failed. What now?

Check the menu bar. Failed sessions offer a Retry Transcription action. Click it and Trace re-transcribes from the saved audio.

If retry doesn’t work, get in touch (see Contact below) with the session’s folder name. I don’t need the audio; the folder name and any partial transcript files are usually enough to work out what happened.

Command line (advanced)

Trace has an optional command-line tool, tracecli, for people who live in a terminal and for AI agents that script their work. It reads your past meetings and summaries, and it controls the live app. Everything it reads stays on your Mac, the same as the rest of Trace. If you only want to record and read transcripts, you never need it.

Install it with Homebrew:

brew install trace-app-mac/tap/tracecli

Reading past meetings

These work whether or not Trace is running, since they read your session folders directly:

[meeting] is optional and defaults to the most recent. You can also pass a row number from tracecli list, or part of a meeting’s name.

Controlling the app

These trigger the same actions as the menu bar and the keyboard shortcuts:

These open a trace:// link that the running app handles, so Trace needs to be running. It will start up on its own if it isn’t.

Reacting to a meeting

tracecli watch waits, then hands on each meeting the moment it is ready. It’s how you pass every recording to something else automatically, without a scheduled job that keeps checking.

tracecli watch --exec ~/bin/my-tool

That runs your command once per meeting, with the meeting’s id as its last argument. Your command can then ask for whatever it needs, with tracecli transcript, tracecli summary, or tracecli path. Leave --exec off and it prints one id per line instead, so a pipe works too.

--require chooses what “ready” means, as a comma-separated list where everything listed has to hold:

• transcript (always included) waits for the recording to finish and be transcribed.
• summary also waits for the summary.
• names waits for you to finish naming the speakers.

names is the one worth knowing about. It’s ready once every speaker apart from your own microphone has a name, or once a minute has passed with no naming happening, whichever comes first. So a meeting you name straight away is handed on immediately, one you half-name is handed on shortly after you stop, and one you never touch still comes through rather than waiting forever. Change the wait with --names-timeout.

Add --json and each line becomes a JSON object with the meeting’s name, its gist, and every file path. --pass transcript sends the transcript markdown to your command’s input instead of its id. --once exits after one meeting.

A watch forgets everything when it stops. --catch-up makes it remember the last meeting it handled, so restarting it replays whatever finished while it wasn’t running.

For example, to summarise every meeting with your own tool once you’ve named who was in it:

tracecli watch --require names --exec ~/bin/summarise-meeting

To keep that running in the background, add --install:

tracecli watch --require names --catch-up --exec ~/bin/summarise-meeting --install

That sets it up to start when you log in and restart if it stops. tracecli watch --service-status shows whether it’s running, and --uninstall removes it.

If you set this up to run automatically in the background (a launchd agent, rather than a terminal you leave open), give tracecli Full Disk Access first, in System Settings → Privacy & Security → Full Disk Access. Trace’s recordings sit in a protected folder that your terminal can already read but a background job can’t. Without the permission macOS tries to ask you, and a background job has no window to ask in, so it just waits. tracecli watch prints a note explaining this if it is left waiting, so check the job’s log if nothing seems to happen.

You only need to do that once. Upgrading tracecli through Homebrew keeps the permission.

Built for agents

Run tracecli --help, or tracecli help <command>, for the full reference. That’s usually all an AI agent needs. The output is made to be easy to script. Results go to standard output while hints and errors go to standard error, reads take --json, and the tool returns a non-zero exit code when something goes wrong.

Contact

Email: [email protected]

Requesting a feature

Email me with the use case. I prioritise features that lots of people ask for, but I read every message.

Reporting a bug

The fastest way to give me what I need:

1. Open Settings → Privacy & Storage → Copy diagnostics. That puts a small text bundle on your clipboard: app version, macOS version, Mac model, and the last 10 minutes of Trace log entries (home paths and email addresses are redacted automatically). Nothing is sent anywhere; it’s just on your clipboard.
2. Email me with the bundle pasted in, plus:
   • What you did
   • What you expected
   • What actually happened
   • If relevant: the session’s folder name from the menu bar list (I don’t need the audio, just the folder name and any transcript files)