blackstackdev/ai-eyes
Permission-first AI co-presence prototype built for OpenAI Build Week - blackstackdev/ai-eyes
0 stars · 0 forks · 0 watchers
README
AI Eyes AI Eyes is a private, local-first co-presence prototype for inviting named AI agents into a shared screen-and-audio session. The v0.1.0 browser build can capture a user-selected screen or window, request system audio, enable a microphone, invite Codi and Lyra, switch between Quiet, React and Explain modes, pause sensing and clear the session. It does not yet connect to an AI provider or claim semantic understanding.
Run and build instructions are in docs/INSTALL.md. Product scope is in docs/FEATURES.md. How Codex and GPT-5.6 were used Codex, powered by GPT-5.6, was the engineering partner for AI Eyes. Starting from Raiden's product idea, it helped turn the permission-first co-presence concept into a working React interface, implement the MediaDevices capture and session lifecycle, diagnose browser and system-audio behavior, run the release QA pass, and prepare the documentation and Devpost submission. Human judgment stayed in the loop throughout. Raiden directed the product vision, privacy boundaries, interaction model and visual decisions, tested the real capture flow on Windows, and approved each release step. We deliberately did not generate fake agent reactions or claim that v0.1.0 already performs semantic audiovisual understanding. Capability boundary Version 0.1.0 proves permissioned source selection, live preview, named-agent presence controls, visible sensing state, pause/resume and session clearing. It does not connect to an external AI provider yet. Multimodal understanding is the next implementation step. Built by Raiden with Codi. CrabDen Software.