# by2kb

> by2kb is a self-hosted CLI turning Bilibili/YouTube URLs into Markdown transcripts, abstracts, and study notes via local Whisper or cloud ASR.

- **Category:** Knowledge Management
- **Licence:** MIT
- **Latest version:** v0.7.5
- **Open source:** oss
- **Languages:** Python
- **Platforms:** Linux, pip
- **Tags:** Markdown, CLI, Ai Agents, Knowledge Base, Youtube, Hermes Agent, LLM, Obsidian, Self Hosted, Whisper, Agent skills, Bilibili, Asr, Video Transcription

## Description

by2kb is a command-line tool for ingesting video content into a personal knowledge base. Given a Bilibili or YouTube URL (or local media file), it retrieves the transcript — preferring native captions, falling back to audio extraction and ASR via local faster-whisper or cloud Doubao — and produces three Markdown files: a verbatim timestamped transcript, a one-minute decision-oriented abstract, and a structured long-form study guide with knowledge map, claims, evidence, and action items. The tool is designed for agent-first workflows: a Hermes plugin intercepts video URLs shared to Telegram, runs ingestion in the background, and returns the artifacts via the same chat. Users without an agent can run the same pipeline with an OpenAI-compatible API key. All artifacts are stored as portable Markdown and JSON in a local directory (default ~/.by2kb) or an Obsidian vault. The project uses SQLite for job tracking and deduplication, requires Python 3.12+, pipx, ffmpeg, and optionally Deno/Node for YouTube player challenges. No resident service or Docker image is provided; it runs as a local CLI subprocess.

## Links

- [Homepage](https://github.com/charlesmpc/by2kb)
- [GitHub Repository](https://github.com/charlesmpc/by2kb)
- [Releases](https://github.com/Charlesmpc/by2kb/releases)

---
Source: https://selfhost.directory/project/by2kb