Book a demo

Detect · on-device

Why Capifiq uses a companion agent.

The most convincing interview cheating doesn't happen in the browser or in the call. It happens on the machine, in ways a web page or a meeting bot can't see. A temporary, session-only companion agent is the only honest way to close that gap — and Capifiq built it to be present for the interview and gone afterward.

Book a demo See how it works

The architecture

Each layer can only see its own layer.

This isn't a preference. It's how software works. A tool can only observe the layer it runs in — so integrity that stops at the browser or the call has a blind spot the size of the whole operating system.

The limit

A browser sees the browser. A call sees the call.

A browser extension can inspect the browser tab and nothing beyond it. A meeting bot can observe what happens in the video call. Both are useful — and both are blind to everything running natively on the computer beside them. That is precisely where the newest assistance tools choose to live.

The answer

A device-level layer sees the device.

A companion agent runs at the device level for the length of the interview, so the tactics that hide from a browser or a call have nowhere left to hide. When the session ends, the layer is removed. Present when it matters; gone when it doesn't.

The blind spot

What only a device-level layer can surface.

These are the tactics that a browser extension or a meeting bot cannot catch — because none of them live in the browser or the call.

Native overlays

Answer windows drawn on top of the interview by desktop software that lives outside the browser entirely, invisible to anything watching only the tab.

Remote-control sessions

A second person quietly driving the machine from elsewhere, so the answers on screen aren't coming from the candidate at all.

Virtual machines and sandboxes

The real interview running inside a disguised environment, with assistance tucked into the layer the call never sees.

System-level interference

Tampering with the display, the camera path, or the desktop itself — happening below the browser and below the call, where only a device layer reaches.

Capifiq describes what it surfaces, never the recipe for how. New tactics appear constantly; the device layer is what lets coverage keep up with them. See the full picture of what we detect →

The trade-off, neutralized

A device-level layer, without the device-level intrusion.

A companion agent could be heavy-handed. Capifiq's is deliberately the opposite: it does the one job it exists for, and nothing else.

What it is

  • A roughly 30-second, guided setup before the interview.
  • Present only for the scheduled session, then guided off the machine.
  • Runs in ordinary user space — no admin or root privileges.
  • Surfaces findings as sealed, timestamped evidence for a human to review.
  • Candidates are told clearly, in advance, with rights and a review process.

What it never does

  • Lock down, block, or restrict anything on the computer.
  • Run before the interview starts or after it ends.
  • Monitor persistently in the background.
  • Take control of the machine or demand elevated privileges.
  • Decide an outcome — a qualified person always makes the call.

Capifiq surfaces the finding. The interviewer makes the judgement. The agent never accuses automatically, and its evidence is never the sole basis for a decision.

Two layers, one picture

The agent watches the device. The meeting bot watches the call.

Neither is enough alone. Together they cover both the machine the candidate is using and the conversation you are having with them.

The companion agent

The device layer. Surfaces on-device assistance the call can't see — overlays, remote control, virtual machines, display tampering, and new vectors as they appear.

You are here

The meeting bot

The call layer. Joins the interview, records and transcribes it, tracks question cadence, and flags likely AI-generated answers as observations for the interviewer.

See the meeting bot →

Questions

Common questions about the agent.

Why not just use a browser extension or a meeting bot?
Because each of those can only see its own layer. A browser extension inspects the browser; a meeting bot observes the call. Native overlays, custom desktop software, remote-control sessions, and virtual machines run outside both — and only a temporary device-level layer can surface them.
How long is the agent on the candidate's machine?
Only for the scheduled interview. It is a roughly 30-second setup beforehand and is guided off the machine afterward. It does not run in the background before or after the session.
Does the agent take over or lock down the computer?
No. It runs in ordinary user space, never takes admin or root privileges, and never blocks, terminates, or restricts anything. It is an integrity check for the interview, not a lockdown of the machine.
Does the agent decide whether someone is cheating?
No. It surfaces findings as sealed, timestamped evidence. A qualified person on the hiring side reviews them and makes the judgement. The agent never accuses automatically and is never the sole basis for a decision. Candidates have documented rights and a review process.

Get started

See the device layer on your own interviews.

Book a walkthrough and watch what a device-level integrity layer surfaces that a browser or a call cannot.

Book a demo Request a pilot