BiliNote — AI Video Note Generator

Introduction

BiliNote is an AI-powered video note-taking tool that extracts audio, transcribes speech, and generates structured markdown notes from any video URL. It supports multiple platforms and LLM backends, enabling developers and researchers to quickly digest long-form video content.

What BiliNote Does

• Extracts and transcribes audio from video URLs automatically
• Generates structured markdown notes with timestamps and sections
• Supports YouTube, Bilibili, and other major platforms
• Integrates with multiple LLM backends (OpenAI, Ollama, local models)
• Exports notes in Markdown, PDF, and Notion-compatible formats

Architecture Overview

BiliNote operates as a pipeline: a downloader fetches audio from the target URL, a speech-to-text engine (Whisper-based) produces timestamped transcripts, and an LLM summarizer restructures the raw transcript into coherent notes with headings, key points, and action items. Each stage is pluggable.

Self-Hosting & Configuration

• Runs locally with Python 3.10+ and ffmpeg installed
• Configure LLM provider via environment variables or config.yaml
• Optional GPU acceleration for transcription via CUDA
• Docker Compose setup for one-command deployment
• Web UI available on localhost for non-CLI usage

Key Features

• Timestamp-linked notes for quick video navigation
• Customizable note templates and summarization prompts
• Batch processing for playlists and channels
• Local-first design keeps your data private
• Plugin system for additional video platforms

Comparison with Similar Tools

• NoteGPT — cloud-only SaaS; BiliNote runs fully locally
• YouTube Summary with ChatGPT — browser extension only; BiliNote is a full pipeline
• Whisper + manual notes — no automation; BiliNote handles the entire workflow
• Glasp — highlights only; BiliNote generates complete structured notes

FAQ

Q: Which video platforms are supported? A: YouTube, Bilibili, Vimeo, and any platform supported by yt-dlp.

Q: Can I use local LLMs instead of OpenAI? A: Yes. Ollama and any OpenAI-compatible API endpoint are supported.

Q: How accurate is the transcription? A: Uses Whisper models; accuracy depends on audio quality and language.

Q: Does it handle multi-hour videos? A: Yes. Long videos are chunked and processed incrementally.

Sources

• https://github.com/JefferyHcool/BiliNote