Blueturboguy07/cue

What it solves

cue provides a private, AI-powered assistant that floats over your screen during live activities. It solves the problem of needing real-time AI help—such as coding solutions or meeting guidance—without the assistant being visible to others during screen shares or recordings.

How it works

It is an Electron-based application that captures three distinct inputs: the user's screen (via screenshots), the user's microphone, and system audio (the other participants in a meeting). These inputs are processed as follows:

• Audio: Transcribed using either cloud providers (OpenAI, Gemini, Deepgram) or locally via a bundled whisper.cpp runtime.
• Visuals: Screen captures are sent to a chosen LLM provider (OpenAI, Anthropic, Google Gemini, or custom OpenAI-compatible endpoints).
• Invisibility: The app uses OS-level flags (setContentProtection on macOS and WDA_EXCLUDEFROMCAPTURE on Windows) to tell the system to exclude the window from screen capture streams.

Who it’s for

It is designed for individuals who want a real-time AI copilot for studying, accessibility, practicing for interviews, or getting live assistance during coding tasks and professional meetings.

Highlights

• Screen-Share Invisibility: Designed to stay hidden from most screen recording and sharing tools.
• Bring Your Own Key: Supports multiple AI providers (OpenAI, Anthropic, Gemini, Azure) and custom endpoints, ensuring no subscription fees to the app developer.
• Local Transcription: Option to run speech-to-text locally using whisper.cpp for increased privacy.
• Multi-Channel Audio: Separates "You" (mic) and "Them" (system audio) channels to provide context-aware meeting assistance.
• Contextual Help: Includes specialized triggers for solving coding problems, generating follow-up questions, and recapping conversations.