Getting started
Everything you need to set up and use GhostDesk.
1. Install
Download the installer from the Download page and run it. GhostDesk installs to your user directory — no admin required.
2. First launch
GhostDesk starts as a floating overlay pinned to the top-right of your screen. It’s automatically excluded from screen captures for privacy.
3. Start chatting
Click the overlay or press the toggle shortcut to open the chat panel. Type a question or click the mic icon for voice mode.
4. You're ready
AI is built in — no API keys or accounts to configure. Just open the overlay and start asking questions.
Toggle visibility
Show or hide GhostDesk instantly. When hidden, it remains running and excluded from screen captures. When shown, it appears as a floating panel.
Click-through mode
Toggle click-through mode. When enabled, mouse events pass through the overlay to applications below. A visual indicator shows when active.
Quick quit
Immediately closes GhostDesk and cleans up all processes. Use this if you need to fully exit the application.
Customization
All shortcuts can be rebound in Settings → Shortcuts. Shortcuts work globally — they trigger even when GhostDesk doesn't have focus.
Built-in AI
The free plan uses Llama 3.3 70B (via Groq) for chat and Llama 4 Scout (via Groq Vision) for screenshots. Paid plans upgrade to OpenAI’s GPT-4o models for higher-quality responses. Everything is built in — no API keys, no accounts, no configuration.
Streaming responses
Answers stream in token-by-token so you get output as it's generated, not after the full response is complete. Feels instant.
Context-aware
GhostDesk maintains conversation context within a session. Ask follow-up questions, refine answers, or change direction — just like a real conversation.
How it works
Click the mic icon or enable voice mode in settings. GhostDesk captures microphone audio, detects when you start and stop speaking, then transcribes your speech to text.
Silence detection
Smart voice activity detection automatically filters silence — no wake words needed. Only actual speech is sent for transcription, reducing latency and cost.
Transcription
Speech is transcribed quickly and accurately using Deepgram Nova-3 (free) or OpenAI Whisper (paid). The resulting text feeds directly into the AI chat for processing.
Requirements
A working microphone. Voice mode uses built-in transcription — no extra keys or setup needed.
Region capture
Use the screenshot tool to select any region of your screen. The captured image is sent directly for analysis.
Vision processing
Captured images are analyzed by ChatGPT's vision capabilities — it can read code, documents, diagrams, charts, tables, and UI, then provides explanations or analysis.
Requirements
No extra setup — vision analysis is built in and works out of the box.