When the Coding Assessment Is an App, Not a Website

Most guidance about remote coding assessments assumes the assessment is a webpage, and that the limits of a webpage are the limits of the monitoring. That assumption stops holding when an employer sends you an installer instead of a link. HackerRank's Desktop App Mode replaces the browser with a native application that HackerRank's documentation describes as adding "operating system–level monitoring to create a fully locked-down environment." If you have been asked to install one of these before a screen, this article covers what the software does to your machine, what it records, what HackerRank documents as its current limits, and how to decide whether to proceed.

The three modes are a ladder, and the top rung leaves the browser

HackerRank documents three integrity settings, and employers choose between them per test. The distinction matters because most of what candidates read online describes only the first two.

Secure Mode stays inside the browser. HackerRank's test integrity documentation says it "runs tests in a controlled browser environment" and "enforces full-screen mode, blocks copy and paste actions, prevents the use of multiple monitors, and alerts when candidates switch tabs."

Proctor Mode adds analysis on top of that, incorporating "AI-powered features such as screenshot analysis, plagiarism detection, and webcam anomaly detection."

Desktop App Mode is a different category of thing. HackerRank documents it as including "all Proctor Mode capabilities" and adding operating-system-level monitoring. It is not a hardened browser tab. It is native software running with permissions you grant at install time.

That gap is the whole story. A webpage is constrained by the browser's security model — a point covered in more detail in what interview websites can and cannot see about your running apps. A native application is constrained only by the permissions you approve.

What installing it actually involves

HackerRank's candidate documentation for attempting tests using the desktop app lists Windows 10 and 11 and macOS Monterey 12 or above, on both Intel and Apple Silicon. There is no Linux build documented.

On macOS the app requests webcam access, screen recording access, System Events access, and Accessibility access. Those last two are worth reading slowly. Accessibility permission on macOS allows an application to observe and control other applications — it is the same permission category that automation and remote-control tools use. Granting it is a reasonable ask for lockdown software to make, and it is also a meaningful grant. It is not comparable to letting a website use your microphone.

The app also expects a cleared machine. HackerRank's documentation tells candidates to "Close all other applications running on your system before starting the test," and offers an option to close them automatically, with the caveat that "any unsaved data is lost." Save your work before you launch it. That warning is not boilerplate.

Once running, the app enforces full-screen mode, and candidates "can exit full-screen mode only after you complete the test."

What the locked environment blocks

Per HackerRank's Desktop App Mode documentation, the app prevents remote access sharing, blocks screen sharing and screenshots, disables copy-paste "to prevent pasting content from external sources," blocks multiple monitors and screen mirroring, and prevents the test from running on a virtual machine.

It also actively intervenes rather than only recording. The documentation states the app "detects and closes any other programs that candidates try to open during the test," and names the categories it targets: video conferencing, messaging applications, and LLM applications. HackerRank's documentation names Cluely specifically as an example of blocked software.

This is the part candidates most often misread. Browser-based proctoring produces a report that a human reviews afterward. Desktop App Mode intervenes during the test. An application that appears mid-assessment does not generate a debatable log entry; it gets closed.

What it watches while you work

The monitoring is behavioral as well as environmental. HackerRank documents continuous webcam analysis that flags when a candidate is not visible or multiple faces appear, object detection for mobile phones and tablets in the webcam feed, continuous checks for additional displays, and code-editor analysis that "detects patterns where candidates type and delete content."

The newest signal is gaze detection. HackerRank's July 2026 release notes describe it as identifying when candidates repeatedly look away from the screen and then resume typing, on the theory that this suggests consulting an external resource. Two details in that announcement deserve attention: HackerRank describes the signal as reported at medium severity specifically "to reduce false positives," and its own documentation states gaze detection is "currently available in limited availability." It is a new and self-acknowledged imprecise signal, not a settled measurement.

The same release notes describe improved screenshot analysis that "detects a broader range of suspicious applications," including newer AI tools, and an AI fluency evaluation that now analyzes IDE activity rather than only AI chat interactions.

Worth keeping in proportion: a flag is a note in a report, not a verdict, and platforms disclose very little about how employers weigh them. What actually happens when a coding assessment flags you covers that downstream process.

It is no longer only take-home tests

Desktop App Mode has extended to live interviews. HackerRank's documentation for Desktop App Mode in Chakra, its AI interview platform, describes the same controls applied to a live session, blocking "the use of other applications during the interview, including video conferencing and AI assistant tools."

The July 2026 release notes add that the AI interviewer now detects when a candidate connects an additional monitor, shares their screen, or exits full-screen, and pauses the interview until the condition is corrected. A dual-monitor setup that is merely a flag on a take-home becomes an interruption in a live session.

What this means if you use AI interview assistance

Directly: a desktop AI assistant is not a workable route through Desktop App Mode, and it is not a permitted one either.

Control is a desktop-native AI interview assistant for Windows and macOS, and desktop AI assistants are the exact category this software is built to close. HackerRank documents blocking LLM applications by name and terminating unauthorized programs during the test. Treating a locked-down native assessment as something to work around means running software the employer has explicitly prohibited, inside an environment engineered to stop it — and the honest read is that this is a category of assessment where an assistant does not belong.

The useful preparation happens before the assessment, not during it. Practising against the format, rehearsing problems out loud, and knowing your own reasoning well enough not to need a prompt is the only approach that survives a locked environment. The AI coding assessment assistant guide covers that preparation workflow, and what HackerRank proctoring detects and records covers the browser-based modes you are more likely to meet.

Documented limits and open questions

HackerRank documents one current gap in its own product. Its Desktop App Mode page states that "content protection is temporarily unavailable in the HackerRank Desktop App because screen recording requires screenshot permissions that were previously restricted." That is a vendor-disclosed limitation on a specific feature as of this writing, not a general weakness, and it is described as temporary.

Two things remain genuinely unclear from public documentation. HackerRank does not publish how gaze detection performs against candidates who look away for ordinary reasons — thinking, reading a second document the test itself permits, or living with a condition that affects eye movement. And it does not publish how employers are advised to weigh a medium-severity signal against a completed, correct submission.

Both of those are reasons to ask questions before test day rather than after.

Practical steps before you install anything

Ask the recruiter which mode the assessment uses, and whether a desktop install is required. This is a normal logistical question and asking it early costs nothing.

Check your machine against the requirements. If you work on Linux, or on macOS older than Monterey 12, raise it before the invitation expires — there is no documented build for you, and that is the employer's problem to solve, not yours to improvise around.

Consider the device. If your only computer is a work machine, installing software that requests Accessibility and screen-recording permissions may conflict with your employer's device policy. That is a legitimate reason to request an alternative or a different machine.

Prepare the environment concretely: close and save everything, disconnect the second monitor rather than hoping it goes unnoticed, put your phone out of webcam frame, and sit somewhere a second person will not walk through the shot.

Read the consent screen. It is the authoritative description of what that specific employer has enabled, and it takes precedence over any general article, including this one.

The short version

The meaningful shift is not that monitoring got stricter. It is that the assessment stopped being a webpage. Once you install native software with Accessibility and screen-recording permissions, the browser's limits no longer describe your situation, and the software can act during the test rather than only report on it afterward.

Treat the installer as the signal it is: this employer has chosen the strictest setting available, and the assessment should be approached on its own terms. If you want to build the underlying fluency ahead of time, Control offers a free allowance of 5 messages and 2 minutes of voice for new users to practice with before test day — and, plainly, that practice is the part that helps here.