by2kb

by2kb is an open-source tool designed to transform video content into structured, durable knowledge base artifacts. Its primary function is to take a video URL (currently supporting Bilibili and YouTube) or a local media file and generate three specific Markdown outputs: an evidence-preserving raw transcript, a short abstract for quick assessment, and long-form study notes. The tool is architected as "agent-first," meaning it is optimized to be used in conjunction with AI agents (such as Hermes) that can handle media retrieval and summary generation using their own model authentication. However, it also supports standalone usage via an OpenAI-compatible API key. by2kb handles the media processing pipeline, including fetching audio, running Automatic Speech Recognition (ASR) via local faster-whisper or cloud providers like Doubao, and normalizing the transcript. Key features include: 1. **Cost-Efficient Pipeline:** It attempts to retrieve native subtitles first (lowest cost), falls back to audio stream extraction and ASR, and finally full media download if necessary. 2. **Separation of Concerns:** It strictly separates the "raw" transcript (the evidence) from "updated" or "abstract" outputs (AI-generated summaries) to prevent hallucination or data loss. 3. **Customizable Skills:** Users can define reusable Markdown-based processing instructions (skills) for tasks like punctuation cleanup, chapter detection, or specific summary formats. 4. **Integration:** It integrates with IM bots (via Hermes plugin) and local knowledge bases like Obsidian or file systems. by2kb requires Python 3.12+, pipx, and ffmpeg. It is suitable for users who want to archive and study video content without relying on browser extensions or cloud-only services, emphasizing local control and portable Markdown artifacts.