A coding agent on a Mac can inspect files and windows but cannot see a cable, a device label, a circuit board, or the object the user is describing. iPhone Camera Bridge gives that agent a temporary, user-controlled view into the physical scene.
A local phone page and Mac viewer that stream the iPhone camera through WebRTC after a user gesture and permission grant.
The aim. Give an agent temporary visual access to physical objects when the user explicitly opens and permits the phone camera.
01The problem behind the project
An agent working on a Mac can inspect files and screens but cannot see a cable, device, label, or physical build without a camera bridge.
The permission model is part of the product. The phone page must be opened, the user must grant camera access, and the session should remain visible and easy to stop. Those steps prevent a useful diagnostic bridge from quietly becoming remote surveillance.
People using agents for hardware, repair, accessibility, and physical-world troubleshooting may benefit. Anyone or anything in camera view is affected.
02How it took shape
A browser-based phone capture surface, local signaling, WebRTC media transport, and a Mac-side viewer designed for deliberate, temporary sessions.
The prototype uses a browser-based phone capture surface, local signaling, WebRTC transport, and a Mac-side viewer. It proves the camera path while keeping private network details, certificates, and captured imagery outside the public case study.
Josiah identified the missing physical-vision capability, defined the user-controlled permission model, and directed agent-assisted implementation.
A working local prototype demonstrates the phone-to-Mac camera path.
03What the project means now
The project is a small but representative agent interface: it extends perception only when the user deliberately supplies it. That pattern is likely more trustworthy than treating every available sensor as ambient agent context.
It is not a general remote-surveillance system. Network details, certificates, private imagery, and unattended capture are excluded from publication.
A user gesture and visible session boundary are product features, not inconveniences, when an agent gains a camera.
Package the local setup cleanly and add unmistakable session, stop, and capture-state indicators.