Tool reference

read_video, search_video, video_evidence

The default read_video profile is fast: container metadata, streams, chapters, and embedded subtitles. It does not detect scenes, extract frames, or run speech recognition.

Each result keeps source locators, an extraction route, warnings, and explicit gaps. Start with the smallest tool that answers the question; use advanced evidence or research operations only when the primary result is insufficient.