Bilibili RAG
A personal knowledge-base tool that turns videos saved in a Bilibili favorites folder into a searchable, source-tracing archive by transcribing them, embedding the text, and answering questions through retrieval-augmented generation.
Bilibili RAG turns interviews, talks, courses, and other videos saved in a user's Bilibili favorites folder into a personal knowledge base that can be searched and queried with sources traced back to the original video. After logging in with a QR code, it reads a chosen favorites folder, transcribes each video's audio (including multi-part videos, part by part), stores the results locally in SQLite and ChromaDB, and answers questions over that content through retrieval-augmented generation with semantic search.
Processed content or AI-organized notes can be exported as Markdown. The project can run through a local Python environment or Docker Compose, using a DashScope-compatible API for the language model, transcription, and embeddings, and also offers a separate hosted online version billed by credits for people who prefer not to self-host.
An OpenClaw skill for local use is also provided.