watch

Video analysis is enabled by downloading the source, extracting timestamped frames, and retrieving synced transcripts to provide full audiovisual context. This allows you to ask questions about specific moments and get accurate answers without manual viewing.

Video analysis works with both public platform URLs supported by yt-dlp and local video files. Supported local formats include .mp4, .mov, and .mkv, ensuring you can process both web content and personal recordings.

A Whisper API key is only required for transcription when native captions are unavailable. The system prioritizes retrieving existing native captions first, falling back to Groq or OpenAI Whisper APIs only when necessary.

To extract frames and transcripts, you need yt-dlp for video acquisition and ffmpeg for frame extraction. Optional Groq or OpenAI API keys are required only if you need Whisper-based transcription for videos lacking native captions.

Diagnosing UI bugs from screen recordings is a supported use case. By extracting timestamped visual frames and synchronized spoken audio, the system provides the full audiovisual context needed to identify and analyze interface issues.

Long-form videos are handled by auto-scaling frame extraction to the video length. This optimization prevents excessive token usage while still capturing enough visual context to provide accurate summarization and question answering.