Invisible Cue Card for Technical Interviews - OpenCluely Breakdown

Imagine this: you're in an online technical interview. You're sharing your screen in Zoom, and your mind goes blank on a binary tree rotation. Googling on a second monitor is risky, and looking away from the camera immediately raises suspicion with the interviewer.

OpenCluely recently gained popularity on GitHub. It's a desktop assistant that's visible to you on screen but completely hidden from screen recording and screen sharing software.

What the app can do

OpenCluely positions itself as a free alternative to Cluely. It runs as an Electron app that creates an invisible overlay on top of your windows.

The main focus of the app is how it interacts with video conferencing software. When you enable screen sharing in Zoom, Google Meet, Microsoft Teams, or start recording in OBS, the video stream simply doesn't include the assistant window.

Information is delivered in two ways:

• You take a screenshot of the problem, and the model analyzes the image.
• You speak your question into the microphone, and the text is recognized in real time.

The response appears character-by-character in a small floating window or chat as it's being generated.

How invisibility is achieved

The developers leveraged operating system APIs to protect content from capture.

On macOS, this is handled by the NSWindowSharingNone parameter. Windows uses the WDA_EXCLUDEFROMCAPTURE flag. Electron sets these options for the app window, so screen grabbers only see what's underneath the overlay.

Linux is a different story. X11 and Wayland lack a similar window manager flag. The authors honestly warn in the documentation: on Linux, the overlay will be visible to all meeting participants. As a partial solution, they suggest streaming not the entire desktop, but a specific window with your code editor.

Voice and multimodal capabilities without extra steps

For visual content processing, OpenCluely sends screenshots directly to Gemini. There's no intermediate OCR text recognition involved. The multimodal model takes the image with code or problem statement and builds the response directly.

The authors also worked on speech. Azure Speech or local Whisper can be selected as the voice engine.

Local Whisper is configured automatically by a script on first launch. Voice input doesn't cut the audio stream on a strict timer. The algorithm tracks natural pauses in speech, assembling a long phrase into a single request. A separate filter removes Whisper hallucinations when it tries to fabricate non-existent words during complete silence.

Hotkeys for control

To avoid drawing attention with cursor movements, control is keyboard-based:

• Cmd/Ctrl + Shift + S opens the area selection for screenshot.
• Cmd/Ctrl + Shift + V instantly hides or shows all assistant windows.
• Cmd/Ctrl + Shift + I or Alt + A enables click-through mode so the mouse ignores the hint window and clicks on elements underneath it.
• Alt + R starts or stops microphone recording.

Installation and platform specifics

The app requires a Gemini API key to work. You can get one for free in Google AI Studio and enter it in the app settings or in the .env file.

Pre-built installers are available for Windows (as .exe) and Linux (packages .deb and .AppImage).

macOS is a different story. There's no ready-made .dmg file in the releases. The app isn't signed with an Apple certificate, so the built-in Gatekeeper security system blocks it from normal launch. Mac owners will need to build the project from source.

The build process is straightforward. You'll need Node.js and Git installed.

git clone https://github.com/TechyCSR/OpenCluely.git
cd OpenCluely
./setup.sh

The setup.sh script handles all the routine: installs NPM dependencies, sets up a Python virtual environment, downloads the Whisper model, and generates the initial .env.

Is it worth using

Using such software in real interviews is in an ethical gray zone. If the interviewer starts asking clarifying questions about the code or asks you to explain your decisions, a generated hint is unlikely to save you.

The app is well-suited for interview practice, solving LeetCode problems under time constraints, and reducing anxiety about algorithm sections.

Among the project's strengths are open source code, no hidden telemetry, the ability to process voice locally, and careful integration with system APIs.