Desktop Capture App

The Desktop Capture App (LMA Capture Client) is a lightweight native application (macOS and Windows) that streams your microphone, your computer’s system (meeting) audio, and — optionally — your screen video directly to LMA. Because it captures the operating system’s audio, not a browser tab, it can transcribe meetings you join from a native desktop app (Zoom, Microsoft Teams, Cisco Webex, Slack huddles, phone bridges, …), which the Chrome Extension and Stream Audio options cannot. It adds no bot or extra attendee to the meeting.

Status: macOS and Windows are available today. The app is distributed as source that you build locally with a one-step script — a native app using ScreenCaptureKit (macOS) or WASAPI/WPF (Windows) cannot be cross-compiled by LMA’s Linux build pipeline, and code-signing tools are OS-specific, so building on your own machine is both required and the most trustworthy option.

How it works

Your microphone is transcribed as the meeting owner — the “My Mic” channel (the AGENT channel downstream).
Your computer’s system audio — the remote participants — is the “Meeting Audio” channel (the CALLER channel downstream).
Audio streams to the same secure WebSocket transcriber endpoint the browser options use; the server is unchanged. From there it flows through the identical pipeline (real-time transcription, Meeting Assistant, summaries, knowledge base).
macOS captures system audio via ScreenCaptureKit (loopback) and the mic via AVAudioEngine. Windows captures system audio via WASAPI loopback and the mic via WASAPI capture. On both, the two mono sources are interleaved into 2-channel 16-bit PCM.

On Windows, system-audio (loopback) capture is built into the OS and needs no special permission — there is no equivalent of the macOS Screen Recording prompt. The only OS gate is the microphone privacy setting.

Optional: identify separate speakers

By default each channel gets a single speaker name — everything from system audio is the Meeting Audio label, everything from your mic is the My Mic label. That is enough for a one-to-one call, but not when several people share a channel.

Settings (⚙) → Speaker identification turns on Amazon Transcribe speaker partitioning (diarization), independently per channel:

Setting	Turn it on when
Identify separate speakers in meeting audio	The call has several remote participants
Identify separate speakers on my microphone	Several people share this mic, e.g. a conference room

Distinct voices are appended to the channel’s label as (spk_0), (spk_1), … — for example Other participants (spk_0). The transcript shows each voice as its own turn, and the labels also reach the meeting summary and Meeting Assistant.

Both are off by default and take effect on the next recording (they ride the session’s opening message, so they cannot change mid-meeting; Settings is disabled while streaming for the same reason). Speakers are numbered per channel, each starting at spk_0, and accuracy is best with five or fewer voices per channel. Labels are numbers, not names: Transcribe distinguishes voices but does not identify whose they are. There is no additional Amazon Transcribe charge.

For headless/CLI use, the same options are --diarize-system / --diarize-mic (or LMA_DIARIZE_SYSTEM / LMA_DIARIZE_MIC).

Choosing the transcription engine

If the deployment also runs the on-demand ASR & diarization engine, --asr-engine microvm sends this meeting there instead of to Amazon Transcribe (LMA_ASR_ENGINE, or asrEngine in lma-config.json). Omit it and the deployment’s own default applies.

Both engines produce speaker labels, so the two flags above do not pick an engine — this is the only way a client chooses one. The on-demand engine separates voices sharing a single microphone more reliably, but it is English only and does not support content redaction, custom vocabularies, custom language models or language identification. A meeting whose MicroVM cannot start falls back to Amazon Transcribe on its own.

Optional: record screen video

In addition to audio, the app can capture and stream desktop video — a screen or a specific window — so LMA saves a video recording of the meeting alongside the audio, just like the Virtual Participant does. It’s off by default; turn it on per user in Settings (⚙):

macOS encodes the chosen display/window with ScreenCaptureKit + VideoToolbox (H.264). It reuses the Screen Recording permission the app already needs for system audio — no new prompt.
Windows encodes with Windows.Graphics.Capture + Media Foundation (H.264). No extra permission is required.

The video is streamed to the transcriber over a second WebSocket connection (so it never delays real-time audio/transcription), muxed with the meeting audio into a single MP4 at the end of the call, saved to the LMA recordings bucket, and shown in the call detail page’s Recording Video player — the same place Virtual Participant recordings appear. Capture runs at ~5 fps and up to 1080p, which keeps CPU and bandwidth modest while staying legible for slides.

When screen video is enabled, the one-time consent gate and the live recording view both show a “Screen video is ON” notice, since screen contents are more sensitive than audio alone. Video recording can be disabled deployment-wide with the EnableVideoRecording CloudFormation parameter (the server then discards any video a client sends).

Backward compatible. The video stream is a purely additive, second WebSocket announced by a new START_VIDEO message. Clients that don’t send it, and servers that predate it, behave exactly as before — audio streaming is untouched.

You are responsible for complying with the legal, corporate, and ethical restrictions that apply to recording meetings and calls — in many jurisdictions all participants must consent to being recorded. Do not use this app to stream, record, or transcribe calls if otherwise prohibited.

The app shows this disclaimer before your first recording and requires you to agree (the same consent gate as the browser extension and the web UI’s Stream Audio tab). After that:

A one-line reminder — “Ensure all participants have consented to recording” — stays visible next to the Start button; hovering it shows when you agreed and the exact text you agreed to.
Settings (⚙ gear) keeps your consent record: the date/time you agreed and the full disclaimer text, expandable for review.
If the deployment’s disclaimer wording is changed by your administrator, the app asks you to agree to the new text once — your recorded consent covers the text you actually saw.

The text is configurable per deployment via the RecordingDisclaimer CloudFormation parameter, so organizations can substitute their own legal wording; the app picks it up from the lma-config.json baked into the download.

Download and install (macOS)

In the LMA web app, open Meeting Assistant ▸ Sources ▸ Desktop Capture App (Native) and click Download for macOS. The zip is preconfigured for your deployment (endpoint + Cognito settings baked in), so you only sign in with your normal LMA username and password.
Unzip it.
If you don’t already have Apple’s command-line tools, install them once:
Terminal window
```
xcode-select --install
```
In Terminal, cd into the unzipped folder and run the installer with bash (not ./):
Terminal window
```
bash install-macos.sh
```
It clears the macOS download quarantine, checks prerequisites, builds the app, and installs it to /Applications/LMACaptureClient.app. (Terminal is only used to build and install — you won’t run the app from Terminal.)

⚠️ Use bash install-macos.sh, not ./install-macos.sh. On recent macOS, running a freshly downloaded script directly makes the kernel execve() a quarantined file, which Gatekeeper blocks with “Apple could not verify … is free of malware” before the script runs — so its own quarantine-clearing never happens. Running it via bash reads the script as data (no execve on a quarantined file), so it runs and clears the flag from the rest of the folder itself.
Launch it like a normal app. Click its Dock icon (the installer pins it), press ⌘-Space (Spotlight) and type LMACaptureClient, or double-click it in Finder. An LMA item appears in the menu bar (top-right) and the app’s icon appears in the Dock.

Searching for it: in the Dock and menu bar the app is labelled LMA Capture Client (<Stack>), but Spotlight and open -a match the bundle’s filename — LMACaptureClient-<stack> — so search for that (the installer prints the exact name when it finishes).

⚠️ Don’t launch it from Terminal. Always launch via Spotlight, Finder, or open /Applications/LMACaptureClient-<stack>.app — never the binary inside Contents/MacOS. Only launching through macOS gives the app its own privacy identity; running it from Terminal makes macOS attribute Microphone / Screen Recording to Terminal, and system-audio capture silently won’t work.
Approve the Microphone prompt. Then open System Settings ▸ Privacy & Security ▸ Screen Recording, enable LMA Capture Client, and quit and relaunch it (Screen Recording requires a restart to take effect). Screen Recording is what lets macOS capture system/meeting audio, even for audio-only capture.
Left-click the LMA menu-bar item, sign in with your LMA username and password, and click Start. Your meeting appears in the Meetings List with a live transcript.

Tip: use headphones. Otherwise your speakers’ meeting audio can bleed into your microphone and appear faintly on both transcript channels.

Later: to move to a newer LMA release see Updating to a new version; to remove the app see Uninstalling.

”Apple could not verify… is free of malware” (Gatekeeper)

macOS flags files downloaded from a browser, and on recent versions it blocks running a downloaded script directly (./install-macos.sh) — the kernel execve()s a quarantined file, Gatekeeper kills it, and the script’s own quarantine-clearing never gets to run. Run it via bash instead, which isn’t blocked:

bash install-macos.sh

To clear the whole folder up front instead (then either form works):

xattr -dr com.apple.quarantine .
./install-macos.sh

Download and install (Windows)

On Windows, loopback (system) audio capture is built into the OS and needs no special permission — the only OS gate is the microphone privacy toggle.

In the LMA web app, open Meeting Assistant ▸ Sources ▸ Desktop Capture App (Native), choose Windows, and click Download for Windows. The zip is preconfigured for your deployment (endpoint + Cognito settings baked in), so you only sign in with your normal LMA username and password.
Unzip it (right-click ▸ Extract All).
Install the .NET 8 SDK once from dotnet.microsoft.com. No admin is required — it can install into your user folder.
In PowerShell, cd into the unzipped folder and run the build-and-install script:
Terminal window
```
./build-windows.ps1 -SelfContained -Install
```
It builds a standalone app, runs a built-in self-test, installs it to %LOCALAPPDATA%\Programs\LMA Capture Client (<Stack>) (no admin needed), and adds a Start Menu shortcut — so you don’t have to dig into the build output. It ends with INSTALL SUCCEEDED; that banner is the thing to look for.
- Omit -Install to just build (the exe lands in bin\Release\net8.0-windows\win-x64\publish\).
- Add -DesktopShortcut for a desktop icon too.
- Add -ProgramFiles to install machine-wide under %ProgramFiles% instead (prompts for admin to do the copy).
Launch it from the Start Menu (press the Windows key, type LMA Capture Client, Enter). No window opens — the app lives in the system tray, at the bottom-right next to the clock:

Left-click that icon for the controls panel. (Once you start recording, a taskbar button appears too — see Using the system-tray + taskbar app.) If SmartScreen warns about an unrecognized app, choose More info ▸ Run anyway — expected for a locally built, unsigned app.
If Windows blocks microphone access, enable it in Settings ▸ Privacy & security ▸ Microphone (turn on Microphone access and Let desktop apps access your microphone), then restart the app. System/meeting audio needs no permission.
Left-click the LMA tray icon, sign in with your LMA username and password, and click Start. Your meeting appears in the Meetings List with a live transcript.

Tip: use headphones. Otherwise your speakers’ meeting audio can bleed into your microphone and appear faintly on both transcript channels.

Later: to move to a newer LMA release see Updating to a new version; to remove the app see Uninstalling.

”Windows protected your PC” (SmartScreen)

Expected for a locally built, unsigned app. Click More info, then Run anyway. (A future release may ship an Authenticode-signed build to avoid this.)

Launched with no arguments (the normal case), the app lives in two places: an LMA item at the top-right of the menu bar, and an icon in the Dock (the installer pins it there).

Menu-bar item:

Left-click the item for the controls popover: sign in / out, Start / Stop / Pause, mute mic, mute system audio, live per-channel level meters, and Open in LMA (jumps to the live meeting in your browser).
While recording, the icon turns red so it’s obvious at a glance.
Right-click the item for Quit (kept out of the popover so it isn’t confused with Stop).

Dock icon:

While recording, the Dock icon shows a red dot and a REC badge (⏸ while paused), and it clears when you stop.
Right-click the Dock icon for quick controls: Start / Pause / Resume / Stop Recording and Open Control Panel.
Click the Dock icon to open the control panel as a regular window (same controls as the popover).

Why both? On MacBooks with a notch, macOS silently hides menu-bar icons that don’t fit — and starting a recording adds the system’s orange mic indicator to the menu bar, which can push the LMA icon out of view at the exact moment you need it. The Dock icon is always visible, so you always have a way to see recording state and stop the recording. If your menu bar is crowded, you can also ⌘-drag the LMA menu-bar icon further right so it’s among the last to be hidden.

Popover options:

Remember my email — prefills your login next launch (email only; the password is never stored).
Start automatically at login — registers the app as a macOS login item. The installer already placed the app in /Applications, so this works out of the box. You can also manage it in System Settings ▸ General ▸ Login Items.
Elapsed time — while recording, the panel and the menu-bar item show how long the recording has been running (also in the icon’s hover text).
Notifications — macOS notifies you when recording starts and stops; the stop notification opens the meeting in LMA when clicked. You are also notified if a window you chose for screen video closes and capture falls back to the whole screen.
Recent meeting names — the clock button beside the meeting-name field re-uses a recent name (handy for recurring meetings).
What will be captured — a summary box above Start Recording (and again while recording) names the microphone in use and, when screen video is on, which screen or window will be recorded, with a small preview of it. This is deliberately on the main panel rather than buried in Settings: sharing the wrong screen is easy to do and hard to undo.
Settings (⚙ gear, top-right of the popover) — opens a separate, resizable Settings window where you customize how the two channels are labeled in the LMA transcript, which microphone is captured, and screen video. Closing the Settings window brings the main panel back.
- My mic speaker label — defaults to your signed-in email address.
- System speaker label — defaults to “Other participants”.
- Microphone — pick a specific input device, or leave System Default to follow your Sound settings. If a chosen mic is unplugged, recording falls back to the default. Changes are saved immediately and apply to the next recording (the gear is disabled while recording).
- Speaker identification — two independent toggles, both off by default: Identify separate speakers in meeting audio and Identify separate speakers on my microphone. When on, Amazon Transcribe tells apart the individual voices in that channel and appends (spk_0), (spk_1), … to the label above (see Optional: identify separate speakers).
- Also record screen video — off by default. When on, choose from a list of your screens and windows, each with a thumbnail preview, its name as macOS reports it (e.g. DELL U2720Q), and its resolution — so two similar monitors are easy to tell apart. Refresh re-scans and re-captures the previews. The chosen screen/window is recorded with the meeting and saved as a video in LMA (see Optional: record screen video).
  
  Thumbnail previews need macOS 14 or later. On macOS 13 the list still shows each source’s name and resolution.

Running it in the background

The app uses no audio or CPU when idle, so the intended usage is to leave it running and click Start when a meeting begins. To relaunch after quitting, click its Dock icon, press ⌘-Space (Spotlight) and type LMACaptureClient, or run open /Applications/LMACaptureClient-<stack>.app.

Using the system-tray + taskbar app (Windows)

Launched with no arguments (the normal case), the app lives in the system tray — the notification area at the bottom-right, next to the clock — with no window. While recording, it also takes a taskbar button, so a live recording is always visible and always stoppable.

Tray icon:

LMA tray icon: gray when idle, red while recording

Left-click the icon for the controls panel: sign in / out, Start / Stop / Pause, mute mic, mute system audio, live per-channel level meters, and Open in LMA (jumps to the live meeting in your browser).
While recording, the icon turns red so it’s obvious at a glance. The app asks Windows to keep it out of the ▲ overflow so the red icon stays visible; if Windows still hides it, drag it onto the taskbar once.
Right-click the icon for Quit (kept out of the panel so it isn’t confused with Stop).
Want it one click away even when idle? Right-click LMA Capture Client in the Start Menu ▸ More ▸ Pin to taskbar. (Windows 10+ removed the API that would let the installer do this for you.)
Only one copy runs at a time. Opening the app again — from that pinned shortcut, the Start Menu, or the .exe — opens the controls panel of the copy already running instead of adding a second tray icon.

Taskbar button (while recording):

A taskbar button appears when you Start and disappears when you Stop — nothing clutters your taskbar between meetings.
Its icon carries a red dot, and the button glows green while recording / yellow while paused.
Hover it for Pause/Resume and Stop buttons right in the thumbnail preview — no need to open a window.
Right-click it for quick actions: Start / Pause / Resume / Stop Recording and Open Control Panel.
Click it to open the control panel as a regular window (same controls as the tray panel).
Closing or minimizing that window does not stop the recording — it stays in the taskbar and keeps streaming. Recording only stops when you choose Stop.

Why both? Windows 11 tucks tray icons it hasn’t seen before into the ▲ overflow flyout, and can hide them again later — potentially at the exact moment you want to check or stop a recording. The taskbar button can’t be hidden that way, so while you’re recording you always have a visible indicator and a one-click Stop. When nothing is recording the app goes back to tray-only and stays out of your way.

Panel options:

Remember my email — prefills your login next launch (email only; the password is never stored).
Start automatically at login — adds a per-user startup entry that launches the tray app when you sign in; the toggle reflects the real system state.
Elapsed time — while recording, the panel, the tray tooltip, and the taskbar button show how long the recording has been running.
Notifications — a tray balloon appears when recording starts and stops; clicking the stop balloon opens the meeting in LMA. You are also notified if a window you chose for screen video closes and capture falls back to the whole screen.
Recent meeting names — the clock button beside the meeting-name field re-uses a recent name (handy for recurring meetings).
What will be captured — a summary box above Start Recording (and again while recording) names the microphone in use and, when screen video is on, which screen or window will be recorded. This is deliberately on the main panel rather than buried in Settings: sharing the wrong screen is easy to do and hard to undo.
Settings (⚙ gear, top-right of the panel) — opens a separate, resizable Settings window where you customize how the two channels are labeled in the LMA transcript, which microphone is captured, and screen video. Closing the Settings window brings the main panel back. Each label field shows its default in grey; leave the field blank to use it, or type to override.
- My mic speaker label — defaults to your signed-in email address.
- System audio speaker label — defaults to “Other participants”.
- Microphone — pick a specific input device, or leave System Default to follow Windows’ input device setting. If a chosen mic is unplugged, recording falls back to the default. Changes are saved immediately and apply to the next recording (the gear is disabled while recording).
- Speaker identification — two independent checkboxes, both off by default: Identify separate speakers in meeting audio and Identify separate speakers on my microphone. When on, Amazon Transcribe tells apart the individual voices in that channel and appends (spk_0), (spk_1), … to the label above (see Optional: identify separate speakers).
- Also record screen video — off by default. When on, choose from a list of your screens and windows, each with a thumbnail preview, the monitor’s own name (e.g. DELL U2720Q), and its resolution, so two similar monitors are easy to tell apart. The chosen screen/window is recorded with the meeting and saved as a video in LMA (see Optional: record screen video).

The app uses no audio or CPU when idle, so the intended usage is to leave it in the tray and click Start when a meeting begins. To relaunch after quitting, press the Windows key, type LMA Capture Client, and press Enter. If it’s already running, that just brings up its controls panel — you can’t accidentally end up with two.

Developer / headless mode. Any command-line flag runs a headless CLI that streams to stdout with a live VU meter (--cli forces it). --selftest validates the login crypto offline, and --capture-test <seconds> <out.wav> records the exact streamed stereo PCM with no server — useful for confirming ch0/Left = system, ch1/Right = mic by measuring per-channel RMS.

Updating to a new version

The app does not update itself. When your LMA deployment is upgraded to a new release, download the new package and re-run the same install command — it replaces the installed copy in place. There is no need to uninstall first.

1. Check which version you have. The controls panel footer and the Settings (⚙) window show a line like LMA Capture Client v0.3.6 · stack LMA-Bob. Compare it with the version shown on the download button in the web app (Meeting Assistant ▸ Sources ▸ Desktop Capture App (Native) — e.g. Download for macOS (0.3.7)). If they match, you’re current.

2. Download the new package from that same page. It is re-generated for your deployment on every LMA upgrade, so it always carries the current lma-config.json (endpoint, Cognito ids, recording disclaimer).

3. Unzip it into a new folder — don’t extract over the old one — and re-run the installer from the new folder:

# macOS — quit the app first (right-click the "LMA" menu-bar item ▸ Quit)
bash install-macos.sh

# Windows — closes the running copy for you, then replaces it
./build-windows.ps1 -SelfContained -Install

On Windows, pass the same flags you used originally — add -ProgramFiles if you installed machine-wide (re-run from an elevated PowerShell), and -DesktopShortcut if you want the desktop icon kept.

What is preserved across an update (as long as it’s the same LMA stack):

Your saved settings — remembered email, speaker labels, microphone choice, screen-video preference, recent meeting names.
Your recording-consent record — you are not re-prompted, unless your administrator changed the disclaimer wording, in which case you agree to the new text once.
macOS permissions — Microphone and Screen Recording stay granted. The installer signs with a persistent local identity, so rebuilds don’t invalidate the grants. (If macOS does re-ask, see Troubleshooting ▸ macOS.)
Windows needs no permission for system audio, and the microphone grant persists.

That’s because everything machine-scoped (bundle id, install path, registry key, login item) is derived from the stack name, which doesn’t change between releases. A package downloaded from a different LMA stack installs alongside the existing one rather than replacing it — see Multiple LMA deployments on one machine.

You can delete the old package folder once the new install succeeds. Keeping it is harmless — its --uninstall / -Uninstall still works for the same stack.

Uninstalling

Both platforms have a single command that removes the app and everything it put on your machine. Run it from any unzipped package folder for that stack (newest is fine):

# macOS
bash install-macos.sh --uninstall

# Windows
./build-windows.ps1 -Uninstall

On Windows you can equally uninstall from Settings ▸ Apps ▸ Installed apps — the entry is LMA Capture Client (Stack) — which does the same thing and doesn’t need the package folder at all.

macOS --uninstall removes: the running app (quit first if open), the LMACaptureClient-<slug>.app bundle from /Applications (and ~/Applications), the Start-at-login item, the Dock pin, saved settings, and the Screen Recording + Microphone privacy grants. It leaves the one-time local signing certificate in your login keychain (harmless, and reused if you reinstall) and leaves the package folder itself — delete that yourself if you’re done.

To remove the certificate too: Keychain Access ▸ login ▸ My Certificates, delete LMA Capture Client Local Signing.

Windows -Uninstall removes: the running app, the install folder (%LOCALAPPDATA%\Programs\LMA Capture Client (<Stack>) and the %ProgramFiles% copy if present), the Start Menu and Desktop shortcuts, the per-user settings key (HKCU\Software\AmazonLMA\CaptureClient\<slug>), the start-at-login entry, and the Apps & features registration. If you installed machine-wide with -ProgramFiles, run it from an elevated (admin) PowerShell — otherwise it warns that it couldn’t remove the %ProgramFiles% copy.

Uninstalling only affects this machine. Meetings, transcripts, and recordings already in LMA are untouched.

If you have clients for several stacks installed, each has its own app/entry and its own uninstall — run the command from that stack’s package folder (or pick that stack’s entry in Apps & features) to remove just it.

Multiple LMA deployments on one machine

If you work with more than one LMA stack, you can install one client per stack and run them side by side. Each download is stamped with the stack it came from, and the app namespaces everything machine-scoped by that name:

What	macOS	Windows
App / install name	`LMACaptureClient-<slug>.app` (shown as `LMA Capture Client (<Stack>)`)	`LMA Capture Client (<Stack>)`
Identity	bundle id `com.amazon.lma.captureclient.<slug>`	registry `HKCU\Software\AmazonLMA\CaptureClient\<slug>`
Settings + consent record	per-stack preferences	per-stack registry key
Start at login	separate login item	separate `Run` entry
Single instance	separate app	separate mutex, so both can run at once
OS permissions	separate Screen Recording / Microphone grants	n/a (loopback needs none)

The stack name is shown under the app title, in the window titles, in the tray / menu-bar tooltip, and in the About line at the bottom of the panel (alongside the app version) — so it’s always clear which deployment you are recording into.

<slug> is the stack name lowercased with non-alphanumerics replaced by dashes, plus a short hash of the exact name — e.g. LMA-VideoTest becomes lma-videotest-4581d5. The hash is what keeps stacks whose names differ only in case or punctuation (LMA-Bob and lma-bob are both valid CloudFormation names) from sharing settings, permissions, or the single-instance lock.