This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
Vision Assistant: Ghost in the Machine — Local Desktop Spectral Agent
What I Built
Vision Assistant: Ghost Edition is a local-first desktop agent that "possesses" your screen workspace. Powered by open-weight AI models, it watches your desktop via global hotkeys, analyzing dense screens, terminal logs, and codebases—responding with eerie, hyper-intelligent insights while running entirely offline on your machine.
Who gets to be a little lazier?
Developers pulling late-night debugging sessions who need a second pair of eyes on their terminal or visual code layouts without uploading personal screen data to proprietary cloud servers.
Demo
- GitHub Repository: Vision Assistant Repo
- Features: Local vision processing, system-wide hotkey trigger, background agent reasoning.
Code
{ https://github.com/chhavigupta783/vission-assisstant-v1}
How I Built It
- Desktop System Shell: Built with Electron & C++ system modules to manage low-latency global screen cropping and desktop triggers.
- Open AI Inference: Powered by open-weight models (Gemma / Ollama) running locally for complete data privacy and offline capability.
- Workflow Orchestration: Managed via n8n local webhooks to pipe desktop context into specialized agent prompt templates.
Why Does Open Innovation Matter?
A true "ghost in the machine" shouldn't phone home to a proprietary server. Running open-weight models locally ensures that sensitive desktop contents, logs, and private code remain 100% on-device. Open innovation allows us to build dark-mode, offline-capable AI tools that work even when the internet is dead.
Prize Categories
- Main Challenge: Ghost in the Machine
- Best Use of Gemma (or relevant partner tag)
Top comments (0)