What is bilinote?

BiliNote is an open-source tool that watches videos on your behalf and turns them into structured written notes. You give it a link from Bilibili, YouTube, TikTok, or Kuaishou, and it downloads the audio, converts speech to text, then passes that transcript to an AI model which produces a clean Markdown document summarizing what was said. The transcription step can be handled by several different speech-to-text engines. The project supports a locally run model called Fast-Whisper, a smaller variant called MLX-Whisper for Apple Silicon machines, and cloud-based options through Groq. For Bilibili videos specifically, the tool can pull the existing subtitle track directly instead of re-transcribing, which saves time. On the AI summarization side, you can plug in whichever large language model you prefer: OpenAI's GPT family, DeepSeek, Qwen, or any other OpenAI-compatible endpoint. The output notes can include screenshots pulled automatically from the video at relevant moments, and each section can carry a link back to the exact timestamp in the original video so you can jump to the source clip. You can also choose a note style: something like an academic summary, a conversational recap, or a key-points-only list. Once a note is generated, you can ask follow-up questions about it using a built-in AI chat that searches the note content to find relevant passages before answering. Deployment is straightforward. A prebuilt Docker image lets you run the entire stack locally with two commands. There is also a source-code path for those who want to modify things, which starts a FastAPI backend and a React frontend separately. For desktop use, Windows and macOS installers are available from the releases page. A browser extension for Chrome, Edge, and Firefox adds a floating button to video pages so you can trigger note generation without leaving the tab. The project is under the MIT license. The README is written primarily in Chinese, reflecting its original focus on the Bilibili platform.