Summary

This project tackles the common issue of personal knowledge systems becoming unmanageable "graveyards" by offering an LLM-compiled knowledge management system. It transforms raw, unedited material into structured, interconnected knowledge like summaries and concepts, bridging the gap between information collection and actionable output. Designed for users seeking to synthesize and leverage their knowledge effectively through AI-assisted compilation.

How It Works

It uses a four-layer, workflow-oriented architecture: raw/ for original, read-only material; wiki/ for LLM-compiled knowledge (summaries, concepts, indexes); brainstorming/ for exploration; and artifacts/ for finished works. LLMs process raw files to generate summaries, extract concepts appearing in multiple sources, and update indexes. A key feature distinguishes "personal practice" from "external viewpoints" within concept entries, highlighting "tension and gaps" for deeper understanding.

Quick Start & Requirements

Download the repo and place test data in raw/. Use Claude Code and run /init-llm for interactive setup. To compile, place an article in raw/articles/ and run /compile. The system supports Claude Code with specific slash commands (e.g., /init-llm, /compile) and Codex, which requires adherence to agent rules and documentation (AGENTS.md).

Highlighted Details

• Automated LLM compilation transforms raw documents into structured summaries and cross-referenced concepts.
• Concepts are automatically extracted and linked if they appear in multiple sources, facilitating synthesis.
• Concept entries clearly delineate "personal practice" from "external viewpoints," surfacing contradictions and knowledge gaps.
• The system supports a workflow-driven approach, allowing dynamic information flow between its four management layers.
• Dual support for Claude Code and Codex offers flexible interaction methods.

Maintenance & Community

Authored by @gatelynch, with original implementation by @claude. No specific community links or sponsorship details are provided in the README.

Licensing & Compatibility

Released under the MIT license, which is permissive and generally allows for commercial use and integration into closed-source projects.

Limitations & Caveats

Functionality depends heavily on specific LLM tools (Claude Code or Codex) and their interaction paradigms. The "friction" of knowledge management shifts from manual organization to conversational clarification, requiring users to address AI-identified contradictions. The project emphasizes a "durable knowledge flow" with evolving tools and workflows, suggesting ongoing development.