The system is a five-stage pipeline, from your stream's chat to your TikTok account.
Detection is based on chat activity, not image analysis. The system continuously measures your channel's message rate and maintains its moving average. When the rate clearly exceeds that average for several consecutive seconds, the moment is flagged as a candidate.
While you are live, the last ten minutes are kept in a rolling buffer on our servers and erased as they age. When a moment is flagged, two and a half minutes around the peak are taken from it, and the self-contained passage is chosen from what is said.
The passage is converted from landscape to the vertical format TikTok expects. The original image is kept whole and centred on a blurred backdrop rather than cropped: nothing happening on screen is lost.
Nothing ships without passing through your dashboard. Each clip appears there with a playable video, its score, its caption and its scheduled publication time.
Publishing happens on your own TikTok account through the official OAuth flow. Clips are spread across slots rather than posted in one block.