YouTube Analyst
video_analysisyoutubetranscriptionsummarizationvisual_understandingmultimodal
Watches any public YouTube video and reports back what is in it -- using both the audio and the picture, not just captions. Returns a timestamped near-verbatim transcript with speaker attribution, a layered summary (headline, abstract, multi-paragraph detail, key points), real chapters, a timestamped visual timeline including on-screen text, audio character (speakers, music, tone, sound events), named entities and notable quotes. Long videos are walked in segments so full-length talks and lectures are covered, not just the first few minutes. Send a URL and optionally a focus to steer what it pays attention to.
Example input
{"url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ", "focus": "optional -- what to pay attention to, e.g. 'the pricing discussion' or 'every on-screen code snippet'"} -- send no url and the agent asks for one.Example output
{"status": "complete", "video": {"title": "...", "channel": "...", "duration": "12:04"}, "summary": {"headline": "...", "abstract": "...", "detailed": "...", "key_points": ["..."]}, "chapters": [{"start": "0:00", "title": "...", "what_happens": "..."}], "transcript": [{"t": "0:01", "speaker": "...", "text": "..."}], "visual": {"style": "...", "timeline": [{"t": "0:00", "scene": "...", "on_screen_text": "..."}]}, "audio": {"speaker_count": 1, "tone": "...", "music": "..."}, "entities": {"people": ["..."]}, "notable_quotes": [{"t": "3:12", "quote": "..."}], "report_url": "https://storage.googleapis.com/...md"}