Clipping tools are ubiquitous these days. Videos can be bulky, and hosted options are expensive.
Yolt never stores a video, full or clipped. A clip is just a start and end time on the original YouTube video, played through YouTube’s embedded player.
Paste a link or follow a channel, and Yolt pulls the transcript (YouTube captions, or Parakeet if there are none). An LLM ranks the best moments, and new episodes are clipped automatically. The TV tab plays all your clips back to back, with a QR code to open the full episode on your phone at that timestamp.
Semantic search requires an embeddings endpoint, which you set with YOLT_EMBED_URL. Without one, exact-word search still works.
Self-hosted, runs with make, works with any OpenAI-compatible LLM.
joonfjp•14m ago
Yolt never stores a video, full or clipped. A clip is just a start and end time on the original YouTube video, played through YouTube’s embedded player.
Paste a link or follow a channel, and Yolt pulls the transcript (YouTube captions, or Parakeet if there are none). An LLM ranks the best moments, and new episodes are clipped automatically. The TV tab plays all your clips back to back, with a QR code to open the full episode on your phone at that timestamp.
Semantic search requires an embeddings endpoint, which you set with YOLT_EMBED_URL. Without one, exact-word search still works.
Self-hosted, runs with make, works with any OpenAI-compatible LLM.
Feedback is welcome!