An EditingAssistant for any OpenAI-compatible chat endpoint (v0.4). Sends the
instruction, the project description and the action guide; expects JSON {reply, actions}
back. The editor validates every action before applying (one undo step).
import { createChatAssistant } from '@video-ultra/ai';
<VideoEditor assistant={createChatAssistant({ endpoint: '/api/assistant' })} />
@video-ultra/ai — optional AI for video-ultra. AI editing (v0.4):
createChatAssistantconnects the editor to your own model endpoint; text to speech:createSpeechEndpoint; captions in any language from your own speech service:createCaptionsEndpoint. Background removal (v0.3): a segmentation model runs in a worker on the device (single-threaded WebAssembly, no cross-origin isolation needed); nothing is uploaded. The model downloads once, the first time someone removes a background, and the browser keeps it.