Live video

Frames in, as they happen.

Your camera, your screen or a device's camera channel is a source of the agent's session. Frames go in as they are made; old frames leave the conversation, so the cost does not grow.

Where frames come from

Three sources, one session.

A session's video is a camera of the page, the page's screen, or a unit's video channel, such as the arm cell's camera in a simulated world.

How live video reaches the agentA frame comes from the page's camera, from a shared screen, or from a unit's video channel, which in a simulation is drawn by this computer's host. The host sends each frame into the realtime session as it is made and deletes old frames from the conversation, so a long session does not resend them. The agent answers in the thread, and the session's usage counts the frames.A camerathe page'sA screenshared from the pageA unit's video channela device, or drawn by the hostThe realtime sessionold frames leave itThe agent answersin the thread; frames countedHow live video reaches the agentA frame comes from the page's camera, from a shared screen, or from a unit's video channel, which in a simulation is drawn by this computer's host. The host sends each frame into the realtime session as it is made and deletes old frames from the conversation, so a long session does not resend them. The agent answers in the thread, and the session's usage counts the frames.A camerathe page'sA screenshared from the pageA unit's video channela device, or drawn by the hostThe realtime sessionold frames leave itThe agent answersin the thread; frames counted

What it saw, in its own words.

The live test puts a picture in front of the camera and asks about it. The operator's name is left out.

test-results/realtime-host/2026-10-15/chat.jsonlreal records · a live session · 2026-10-15
22:23:57.764 note    person:operator: What do you see on my camera? Name the colour of the shape and the word written on it.
22:23:58.542 note    agent:openai-gpt-realtime-2@livebox: I see a red square with the word "CommandAGI" written on it in white.

Case study · the camera, live

Two frames, one right answer.

OpenAI's gpt-realtime-2 answered from the frames the session sent, and the session's usage record counts them.

What is not shown. The screen and a unit's camera channel are built and held by browser journeys and tests (a simulated arm cell's camera, drawn by the host). This page has no live measured session of them yet.

frames sent
2
from the question to the end of the answer
2.3 s
for the session
$0.007
ProvenanceFile test-results/realtime-host/2026-10-15/summary.json (this computer)Output case-studies/realtime/stats.jsonTool the live realtime test, read by scripts/sites/case-studies.mjs realtimeRan on This computer (dell-gb10-1), OpenAI's realtime APIWhen

Show your agent what you see.

Start a realtime session in a thread and turn on the camera or share the screen.