What it does BiliNote is an open-source AI video note generator. Paste a link from Bilibili, YouTube, Douyin, or Kuaishou and it extracts audio (or grabs subtitles directly), transcribes via Whisper or Groq, then asks your chosen LLM to produce structured Markdown notes. The output includes optional auto-captured screenshots, clickable timestamps back to the original video, and a cover banner. There’s also a RAG-based Q&A mode where you can interrogate the note contents with function-calling support.

The interesting bit The project is aggressively multi-modal and multi-platform in ways that suggest actual dogfooding. It tries subtitles before audio download when available (including Bilibili’s own subtitles via browser cookie), falls back to local Whisper with model corruption self-healing, and ships as a web app, desktop client (Tauri), browser extension, and Docker image—with a hosted Pro version for those who don’t want to wrestle with proxy configs and model downloads.

Key highlights

• Supports Bilibili, YouTube, Douyin, Kuaishou, and local video files
• Multiple transcription backends: Fast-Whisper, MLX-Whisper (Apple Silicon only), Groq API, BCut
• Pluggable LLM providers: OpenAI, DeepSeek, Qwen, etc.
• Browser extension (Chrome/Edge/Firefox) with sidebar viewer, popup, and floating action button
• Desktop app with first-run wizard, backend health monitoring, and auto-cleanup on exit
• Docker deployment with explicit volume guidance (the README warns strongly against mounting the entire backend directory)
• RAG-based AI Q&A with function calling for querying source material

Caveats

• Windows desktop users must avoid Chinese characters in the install path
• Local Whisper model downloads can trigger OOM on first run; default is now tiny (~75MB) rather than medium
• The README is Chinese-first; English documentation appears minimal or absent
• Several recent patch releases were CI/build fixes (pnpm version pinning, Node compatibility, missing lockfiles) suggesting the build pipeline has been temperamental

Verdict Worth a look if you consume long-form video content in Chinese or English and want searchable, quotable notes. The hosted Pro version removes the deployment friction that the README spends considerable space troubleshooting. Skip if you need mature i18n, a stable self-hosted experience without Docker wrangling, or primarily non-Chinese video platforms with poor subtitle availability.