I've been frustrated with the current state of AI assistants. We're always switching tabs to ChatGPT or Claude, copying and pasting context back and forth. It feels like we're working for the AI, not the other way around.
It's a system-wide, context-aware text completion tool that uses Google's Gemini 1.5 Flash to take screenshots of your active window. This gives it visual context, so it understands whether you're in your IDE, a terminal, a browser, or word processor, and provides more relevant suggestions.
The vision is to make AI assistance as seamless as autocorrect, but infinitely smarter—a "system cursor" that works everywhere.
This is where I need the open source community's help.
Right now it's X11-only on Linux, but the real potential lies in making this work everywhere. I'm looking for contributors to help with:
Windows implementation (Win32 APIs)
macOS support (Accessibility APIs)
Wayland compatibility
Local model integration (Ollama, GPT4All)
Better OCR and context extraction
Performance optimizations
This is a seed-level experiment that could fundamentally change how we interact with AI—but only if the community helps it grow. The future of contextual AI shouldn't be locked behind proprietary walls.
What would it take for you to use something like this? And more importantly want to help build it?
teen-different•7mo ago
So, I built an experiment to flip that script: an AI that follows you. https://github.com/Pi4Wear/systemcursor
It's a system-wide, context-aware text completion tool that uses Google's Gemini 1.5 Flash to take screenshots of your active window. This gives it visual context, so it understands whether you're in your IDE, a terminal, a browser, or word processor, and provides more relevant suggestions.
The vision is to make AI assistance as seamless as autocorrect, but infinitely smarter—a "system cursor" that works everywhere.
This is where I need the open source community's help. Right now it's X11-only on Linux, but the real potential lies in making this work everywhere. I'm looking for contributors to help with:
Windows implementation (Win32 APIs) macOS support (Accessibility APIs) Wayland compatibility Local model integration (Ollama, GPT4All) Better OCR and context extraction Performance optimizations
This is a seed-level experiment that could fundamentally change how we interact with AI—but only if the community helps it grow. The future of contextual AI shouldn't be locked behind proprietary walls.
What would it take for you to use something like this? And more importantly want to help build it?