Paste a YouTube link and get a summary, using a local model via Ollama.
boxslime detects whether a link is a single video or a playlist, fetches transcripts with a bundled yt-dlp (downloaded once on first run), and summarizes them locally.
Long transcripts are split into chunks, each summarized, then synthesized into one summary (map-reduce) so they fit a small model's context. Results stream to the UI, so a playlist fills in video by video.