Windows — rollout, headset, and capture setup (optional)
Windows
This page covers live call audio for Agent Assist on Windows: headset and sound settings, microphone access, network, and pilot validation. It is optional Step 4 after Screen Recorder is deployed on Windows PCs.
Before call audio
Complete install and baseline security on Screen Recording Requirements, then return here:
Step 1: Install · Step 2: Security policy · Step 3: Launch the app · Agent Assist firewall destinations
LevelAI enables live call audio per tenant and can assign desktop capture to selected agents. Those agents use Level Screen Recorder for the agent and customer audio legs. Other agents continue on CCaaS audio for transcription. In both cases, the CCaaS integration supplies call session events (call start, hold, transfer, and call end) that trigger capture. See Trigger-based recording.
During an active call, Level Screen Recorder captures two separate audio legs and sends them to Agent Assist for transcription and in-call guidance:
| Leg | What is captured | Windows source |
|---|---|---|
| Agent | The agent’s speech | Default Communications microphone (headset microphone) |
| Customer | What the agent hears from the caller | Loopback of default Communications playback (headset earpiece / CCaaS output) |
Call audio is independent of the screen-recording stream. Screen recording may be off while call audio continues, when the agent is entitled for desktop capture.
Agent workflow: Sign in and work in the Agent Assist workspace with the assigned headset and CCaaS client. Capture starts and stops with the CCaaS call. Open Recording On or Recording Off in the title bar. The Audio row shows OFF when idle and ON while capture is running.
Platform: Windows 10 or Windows 11 (64-bit), interactive desktop on a physical PC. Virtual desktops are covered in Virtual desktops.
Level Screen Recorder does not capture call audio continuously. Capture starts and stops when the CCaaS integration sends call session events to Level. The integration controls when capture runs. It does not replace the audio captured on the agent PC.
Confirm these events with LevelAI before rollout:
| Call session event | What it does |
|---|---|
| Call start | Starts capture for that call. |
| Hold | Updates the call session while the call is on hold. |
| Transfer | Updates the call session when the call is transferred. |
| Call end | Stops capture for that call. |
Complete the testing cycle with LevelAI before production. The cycle validates the path from each CCaaS event to capture on the agent PC and to the transcript in Agent Assist. A single connected call is not enough. Include hold and transfer.
Confirm the following before rollout (install and baseline firewall are in Screen Recording Requirements):
Complete every item with LevelAI and your pilot team before production. Resolve any No or Unsure answers, then export a report.
The sections below are the configuration reference for each check.
| Requirement | Detail |
|---|---|
| OS | Windows 10 or Windows 11, 64-bit (x64). |
| Session | Interactive desktop (the agent is signed in at the console). Session 0 and service-only sessions cannot capture call audio. |
| Headset | One device for both microphone and earpiece on the call. Wired USB headsets are preferred. |
| Windows sound | That headset is Default Communications Device on the Recording tab and the Playback tab. |
| CCaaS audio | Softphone or browser uses the same communications playback and recording devices. |
| Microphone privacy | Windows allows Level Screen Recorder to use the microphone. |
| Network | Outbound HTTPS and WSS (TCP 443) as listed below, for the signed-in user. |
| Endpoint security | Capture helper (gst-launch-1.0.exe under the Level installation folder) is allowed to run. |
Call audio uses the Windows communications microphone and a loopback of communications playback. A second sound card, Stereo Mix, or a virtual cable is not required.
Level uses the Windows Default Communications Device. This is separate from Default Device, which Windows uses for music and system sounds.
Configure the CCaaS headset as the communications device for both recording and playback. If communications playback remains on laptop speakers while the agent listens on a headset, customer audio is not captured. If communications recording remains on the laptop microphone array while the agent speaks into the headset, agent audio is not captured.
Complete these steps on every agent PC after connecting the headset, and again after Windows resets the default (for example after unplug, dock, or Bluetooth reconnect).
Windows 11: Settings → System → Sound → More sound settings. This opens the Sound control panel with Recording and Playback tabs and the communications default.
| Type | Recommendation |
|---|---|
| Wired USB headset | Preferred. Connect before the shift. Most consistent for both legs. |
| USB handset / desk phone dongle | Supported when Windows exposes it as the communications recording and playback device, and CCaaS uses it. |
| 3.5 mm analog headset | Supported when it is the communications default. On combined analog jacks, confirm both input and output. |
| Bluetooth headset | Supported when it remains the communications default. Pair and connect before sign-in. Keep the same profile (hands-free) for the duration of the call. After reconnect, set the communications default again if Windows changed it. |
| Laptop speakers and built-in microphone | Use only when they are the intended call path. When the agent uses a headset, set that headset as the communications device so capture matches what the agent hears and speaks. |
After disconnect: Windows often restores the laptop as the communications default. Capture resumes after the headset is again the communications device. Level does not substitute laptop speakers or the built-in array for an active headset call path. Reconnect the headset, restore communications defaults, then continue.
Exclusive mode: Some softphones take exclusive control of the headset. Level uses shared mode. If the banner Audio device not available appears while the headset is connected, open Sound control panel → device → Properties → Advanced, and clear Allow applications to take exclusive control of this device.
Multiple devices: Docking stations, webcam microphones, and “Listen to this device” add extra endpoints. Leave unused devices disabled, or leave them unset as the communications default.
Customer audio is the signal that plays on the communications playback device. Route the call through that device.
The agent leg requires Windows microphone permission for Level Screen Recorder. The customer leg uses playback loopback and does not replace this permission.
Intune / Group Policy: Deploy an allow rule for Level Screen Recorder (the desktop app). A policy that blocks desktop applications from using the microphone replaces Agent Assist with the audio permission recovery screen.
When permission is missing, Agent Assist is replaced by a recovery screen titled Audio recording permission is needed for Levelai app to work. Agents select Enable Audio (opens Windows microphone settings) and follow the on-screen steps:

The Recording menu Audio row also shows an alert and Enable Audio. After microphone access is granted, Agent Assist returns and Audio shows OFF until a call starts.
Agents who use Agent Assist without desktop call audio do not require this additional microphone policy.
Allow the required destinations and Agent Assist firewall destinations in Firewall Whitelisting on Screen Recording Requirements, for both the signed-in user and SYSTEM.
Call audio also requires:
| Destination | Port | Purpose |
|---|---|---|
agent-assist.thelevel.ai |
TCP 443 (HTTPS and WSS) | Live call-audio stream (two secure WebSocket connections per call — agent leg and customer leg) |
screen-prod.thelevel.ai |
TCP 443 (HTTPS, long-lived) | Same Trigger host as recording controls. The application maintains a connection so capture can start and stop with the CCaaS call. |
aa-auth.thelevel.ai |
TCP 443 | Listed in the Agent Assist table. Used for session and call-audio entitlement. |
Apply the same HTTPS inspection guidance on Screen Recording Requirements. Inspection that interrupts WebSocket upgrades to agent-assist.thelevel.ai can prevent capture from reaching ON during a call.
Dedicated or tenant-specific Level environments may use different hostnames. Use the list Level provides for the rollout.
Call audio starts a bundled helper, gst-launch-1.0.exe, from the Level Screen Recorder installation folder (in addition to helpers used for screen recording). Allow this process to run for the duration of the call.
Extend the allowlists from Step 2: Security policy:
gst-launch-1.0.exe and related files under %ProgramFiles%\Level Screen Recorder\ when launched by Level Screen Recorder.agent-assist.thelevel.ai.Allow EDR to leave helper binaries and DLLs in the Level Screen Recorder installation directory.
Provide this sequence to every agent, or include steps 2–4 in the image or Group Policy.
If Audio stays OFF during a live CCaaS call, the workstation is configured and the call has not been signaled to Level. Contact Level or IT to confirm Agent Assist entitlement and CCaaS integration.
In a standard Watchdog deployment (screen recording session plus Agent Assist), call audio appears in the Recording menu: select Recording On or Recording Off in the title bar. The menu lists Audio and Screen.
| What agents see | When | Agent action |
|---|---|---|
| Agent Assist home | Signed in, not on a captured call | Continue work. |
| Audio — OFF | Between calls | None |
| Audio — ON | Capture is running for the current CCaaS call | Use transcript and guidance in Agent Assist |
| Audio — Reconnecting | Temporary stream retry during a call | Continue the call. Guidance may pause briefly. |
| Audio — alert plus Enable Audio | Windows microphone access is blocked | Select Enable Audio and complete the recovery screen |
| Recovery screen: Audio recording permission is needed for Levelai app to work. | Same microphone block; Agent Assist is hidden until access is granted | Enable Audio, complete Steps 1–2 in Windows Settings, then Enable Audio again in the app |
| Banner: Audio device not available | Communications headset missing, disconnected, or in exclusive use | Reconnect the headset and restore communications defaults |
| Banner: Unable to start | Capture did not start for this call | Continue the call. Escalate if it persists. |
| Banner: Not supported on this computer | Session cannot capture (typical on VDI / non-console) | Contact IT. Use a physical Windows PC unless Level has approved the profile. |
| No Audio row | Desktop call audio not enabled for this agent, or Agent Assist session unavailable | IT confirms entitlement with LevelAI |
Run this cycle on a pilot PC with LevelAI after the call audio rollout checklist items, and before production. It confirms that each CCaaS session event triggers the expected capture and that both sides of the call appear in Agent Assist.
Move to production only after call start, hold, transfer, and call end have each been validated.
On many VDI and DaaS sessions, call audio is unavailable. The Agent Assist banner shows Not supported on this computer, and the Recording menu Audio row shows an alert. Use a physical Windows PC (interactive console session) unless Level has confirmed a VDI audio-redirection profile for the environment.
The Windows VDI section on Screen Recording Requirements covers Recorder on a golden image. That model does not by itself enable call audio on pooled desktops.
agent-assist.thelevel.ai and the Trigger connection on screen-prod.thelevel.ai.When opening a ticket for call audio, include: