Invisible on shared screens. Grounded in your own resume. Ready when the interview turns technical.

Because Wraith validates the device license while you use the app and when actions are performed. That live check needs internet, but your resume, interview context, local Ollama inference, and local Whisper transcription still stay on your machine.

Join Google Meet from a laptop and a phone. Run Wraith on the laptop, share the laptop screen, and watch the shared view from the phone. If Wraith does not appear there, Ghost Capture is behaving correctly in that setup. Use the same pass to verify the microphone, screenshot, and snapping controls.

Wraith uses local Whisper to transcribe whatever reaches the active microphone when microphone transcription is turned on. That transcript is then sent to the selected Ollama model. If the target audio does not reach the active microphone cleanly, transcript quality drops.

No. Both flows are manual. You need to click the microphone icon or use Ctrl+Shift+M when you want Whisper to listen, then give it a short silence so it can finish and turn off, or stop it yourself with the same control. Screen analysis is also manual, using the screenshot control or Ctrl+Shift+S only when you want Wraith to analyze the current screen.

Yes. Wraith can work with different local Ollama models, and different context sizes will affect latency, memory use, and overall responsiveness. Bigger models and larger contexts can demand more GPU memory, more system RAM, and more CPU headroom.

Start with the ₹100 for 24 hours plan unless you already know you need longer access. It exists for interview day, coding rounds, and short bursts of usage so you do not have to jump into a bigger spend by default.

No. This site is for the installer, compatibility warnings, and docs. Every Wraith payment and recharge happens inside the desktop app after install.

Wraith exposes global hotkeys for showing or hiding the window, starting or stopping microphone transcription, capturing the current screen, and snapping the window through its layout grid. The microphone and screenshot flows are manual, not automatic.

Wraith uses a device-bound license instead of an account system. If you move to a new device, reinstall to a different operating system, or change the machine enough to alter its fingerprint, Wraith can treat it like a new device and a new purchase may be required.

That is a deliberate privacy choice. Wraith avoids collecting email addresses and account profiles because holding personal user data creates more privacy risk, more breach exposure, and more compliance overhead. The app only needs a device-bound license to work, so it avoids storing identity data it does not actually need.

Note: When you purchase a plan inside Wraith, the payment is processed through Razorpay. Razorpay may ask for your mobile number as part of their checkout flow. That is Razorpay's own requirement for payment processing, not something Wraith collects or stores.

No, not by default. Resume parsing, local Ollama inference, and local Whisper transcription stay on-device. Wraith being online for license checks does not automatically mean your interview content is being sent to a cloud model.

Wraith is built around local Ollama models and can populate the models installed on your machine. The preferred models are moondream (great for vision tasks) and qwen3.5:0.8b (lightweight and fast). Install them via terminal:

Yes, on Windows, macOS, and Linux.

On macOS: Apple Silicon (M1, M2, M3, M4 and later) uses ARM architecture and Wraith has a dedicated Apple Silicon build for it. Download the arm64.dmg installer for Apple Silicon Macs.

On Windows: The current public Windows release fully supports ARM processors. There is a single universal Windows installer that automatically checks your architecture (x86, x64, or ARM) and installs properly.