Scrolls

Scrolls is a local-first CLI for turning saved and referenced content from many platforms into an agent-readable personal knowledge library.

It is inspired by Field Theory CLI's flow for X/Twitter bookmarks:

1. Sync saved content from a platform.
2. Enrich each item with source metadata, media/link context, transcripts, article text, or page extracts.
3. Classify items by category, domain, concepts, and usefulness.
4. Store every item as structured data plus an individual Markdown scroll.
5. Compile a Karpathy-style interlinked knowledge base.
6. Expose the library to agents such as Claude Code, Codex, and Hermes via CLI commands and skills.

Initial platform targets

• X/Twitter bookmarks, via Field Theory-compatible concepts
• YouTube videos/playlists, including transcripts and channel metadata
• Wikipedia pages, including summaries, page metadata, and linked concepts
• Web pages/articles, including readable text extraction
• GitHub repositories, issues, releases, and README metadata
• PDFs/arXiv papers, where extractable

Working product shape

scrolls sync youtube <url-or-playlist>
scrolls sync wikipedia <page-or-query>
scrolls sync web <url>
scrolls classify
scrolls search "distributed systems"
scrolls show <id>
scrolls md
scrolls kb
scrolls agent install

Library layout

~/.scrolls/
  items/          # normalized JSON/SQLite source records
  scrolls/        # individual Markdown files
  library/        # compiled interlinked KB
  media/          # optional thumbnails, images, transcripts, attachments
  config.toml

Design principles

• Local-first and agent-readable
• Platform adapters over one-off scrapers
• Markdown files as durable artifacts
• SQLite FTS/BM25 for fast local search
• Classification that works with or without an LLM
• Explicit provenance for every saved item
• Useful from shell, coding agents, and Hermes skills

Status

New repository scaffold. Implementation TBD.