AI-generated Markdown notes from your video backlog

What it does

BiliNote ingests links from Bilibili, YouTube, Douyin, Kuaishou, or local files, transcribes the audio via local or remote engines, and uses a configurable LLM to generate structured Markdown notes. The output can include auto-captured screenshots, clickable timestamps back to the original video, and a persistent history of generated versions.

The interesting bit

Its browser extension sidesteps the entire audio pipeline for Bilibili by pulling subtitles directly from the page using your logged-in cookies. The project also packages everything into a desktop app for Windows and macOS, plus a Dockerized web service, with a hosted Pro version at bilinote.app for users who’d rather not manage Whisper downloads and proxy configs.

Key highlights

• Supports multiple LLM backends (OpenAI, DeepSeek, Qwen) and transcription options (Fast-Whisper, MLX-Whisper on Apple Silicon, Groq, BCut), plus an optional multi-modal video understanding mode
• Browser plugin (Chrome/Edge/Firefox) provides popup controls, a floating video-page button, and a sidebar with Markdown, mind-map, and AI chat views
• RAG-based Q&A indexes video metadata and note content, with Function Calling so the model can retrieve specific source data on demand
• Desktop client monitors backend health, validates install paths for Chinese characters or spaces on Windows, and walks through a 4-step first-run setup
• Docker deployment supports GPU acceleration and splits persistence across separate volumes for data, configs, screenshots, and cached models

Caveats

• Windows desktop users must avoid Chinese characters or spaces in installation paths, and Linux desktop builds were dropped from CI due to slow build times; Linux users are directed to Docker instead
• Recent patch releases (v2.2.1 through v2.2.3 and v2.1.2) fixed build failures caused by pnpm/Node version mismatches and missing lockfiles, suggesting the release pipeline has been temperamental
• The README notes that public Docker mirror sources in China are frequently shut down, which may complicate local builds for users in that region

Verdict

Worth a look if you consume long-form video content and want searchable, structured notes without manual scrubbing. Less interesting if you already have a transcription-to-LLM pipeline you like, since BiliNote’s main value is the all-in-one packaging and RAG chat interface.