Loop Camera

Loop Camera observes a camera with an independent vision model and reports according to your condition. The foreground conversation continues; its model does not need image support.

Use it

Select a monitor camera and model in desktop settings, allow camera access, then describe a condition, such as “Watch this scene and let me know when someone appears.” The selected device remains fixed for that task. Ending monitoring releases the camera.

Runtime behavior

Item	Current behavior
Concurrency	One monitor at a time
Default sampling	Every 2.5 seconds; slow processing does not create a catch-up queue
Default duration	30 minutes
Condition hit	Report once and continue; two consecutive misses rearm reporting
Consecutive failures	End after three capture or inference failures
Stop or timeout	Cancel capture and inference, then release the camera

Built-in and USB cameras are selected by device ID. A disconnected selected device does not silently fall back to another camera. Network cameras, RTSP, phone cameras and continuous video context are outside current support.

Compared with conversation vision

Conversation vision captures one frame from the default camera for a submitted transcript or text turn. Loop Camera has its own task, device, sampling interval and model. Images and observations remain evidence; text visible in a frame cannot authorize operations.

See the support matrix for model selection and native vision and independent monitoring for implementation details.