What problem does it solve? Long videos exceed what a single model request can hold, and re-watching a video for every question wastes time and API calls. This Skill builds a persistent, queryable memory of a video's people, dialogue, events, and facts so questions are answered in milliseconds without re-uploading. ## Core Features & Use Cases - Stateful memory building: Slice long videos into 30-second windows and extract people, speaker-attributed utterances, paralinguistics, acoustic events, and semantic facts via an omni model, with resumable and appendable builds. - Hybrid retrieval: Query by person, dialogue content, exact fact keys, time ranges, or open-ended hybrid search across episodic and semantic containers. - Selective replay: Re-watch specific source clips with audio when a detail was not captured in the text record. - Use Case: Given a 2-hour meeting recording, build a memory once, then ask "What did David say about the budget?" or "What did they agree on in the end?" and get answers with speaker attribution and timestamps. ## Quick Start Ask the agent to check the memory status of your video file and build a memory for it, then ask your question about the video's content.
