How do AI interview assistants capture system audio?

In a video interview, there are two distinct audio streams on your machine:

• Microphone input — your own voice, captured by the mic.
• System audio output — everything your computer is playing through the speakers, including the interviewer's voice coming from Zoom, Google Meet, or Microsoft Teams.

An AI interview assistant needs both streams to follow the conversation: the interviewer's questions (system audio) and your answers (microphone). Capturing only one side produces a partial transcript and weaker suggestions.

System audio capture relies on operating-system audio APIs rather than on the meeting app itself. The exact mechanism differs by platform:

• macOS — modern versions expose process and system audio taps through Core Audio. Older approaches used virtual audio devices (loopback drivers) that route the system output back in as an input.
• Windows — the Windows Audio Session API (WASAPI) supports loopback capture, which lets an application record whatever is being played out of a chosen output device.

Either way, the capture happens locally on your device. The assistant does not need to be "inside" Zoom or Teams; it reads the audio after the meeting app has already decoded it for playback. You can read more about the overall pipeline on the SubcueAI homepage or the tutorial.

SubcueAI is a native desktop app for macOS and Windows. It uses dual audio capture: one stream for your microphone and one stream for system audio coming from the meeting app. Both streams are transcribed so the assistant can tell who said what.

• No meeting bot joins the call as a participant.
• Nothing is injected into Zoom, Google Meet, or Teams.
• Suggestions appear in a floating local overlay on your own screen.

Because the overlay is rendered locally, it is not part of the video stream you send to the interviewer. For more on the design choices behind this, see About SubcueAI or how it compares to alternatives.

System-audio capture works on your own personal computer. It does not change what an interviewer can observe in these situations:

• Screen sharing — if you share your entire screen, a local overlay that is a normal window is visible to the interviewer; SubcueAI's desktop overlay is excluded from capture on Windows 10 (version 2004) and later and stays hidden in Google Meet on macOS, though some Mac apps, such as Tencent Meeting and QuickTime, can capture it.
• Screen recording or proctored exams — recording tools can capture an overlay that is not excluded from capture (on a Mac, QuickTime records even an excluded one), and proctoring software can see running processes regardless of how audio is tapped.
• Company-managed or locked-down devices — IT policies may block third-party apps from installing or from accessing audio APIs.
• Headphones-only setups — if the meeting app routes audio to a Bluetooth headset in a way the OS does not expose, loopback capture can be inconsistent.

For more context on what is and is not observable, see Security.