BiliSum
BiliSum generates AI summaries, transcripts and a searchable knowledge base from Bilibili, YouTube and local videos, usable from a desktop app, a command-line tool, or as a skill an AI coding agent can call.
BiliSum transcribes a video, then produces a text summary, an illustrated note that reads the transcript alongside video screenshots, a mind map, and an entry in a cross-video knowledge base that supports semantic search, keyword search and retrieval-augmented question answering. Transcription can use SiliconFlow ASR, other multimodal ASR services, or local Whisper and FunASR models, and the illustrated notes can use a configurable vision model to pull objective facts from candidate video frames and filter out low-quality ones.
A desktop app for Windows and macOS handles Bilibili sign-in by QR code and in-app updates; a command-line tool installed from npm can run independently or alongside the desktop app, sharing the same tasks, configuration and knowledge base, and a bundled skill lets an agent such as Codex, Claude or Cursor choose between a full summary, a quick brief, a transcript or a task-status check. A Docker image is also available.
The desktop client is built with Electron, React and TypeScript, the backend with FastAPI and SQLite, and all data, including transcripts and the knowledge base, is kept on the local machine.