Karpathy Wiki Graph

lcwiki

[English](README.md) | [简体中文](README.zh-CN.md)

[](https://pypi.org/project/lcwiki/) [](https://pypi.org/project/lcwiki/) [](./LICENSE) [](https://github.com/LCccode/wikigraph/actions/workflows/ci.yml)

**The enterprise-grade knowledge base agent, grounded in personal knowledge management theory, built for cross-department KB co-creation.** Type `/lcwiki` in Claude Code or OpenClaw — lcwiki plugs into your AI agent as its **long-term KB brain**, trains on any folder of docs your teams drop in, and answers cross-department queries with **precision knowledge allocation** at ~10% the token cost of vanilla RAG and **~10× lower ongoing training cost** via smart incremental ingest.

**Mount lcwiki on OpenClaw or Claude Code as a drop-in agent brain.** One `pip install`, one `lcwiki install --platform claude|claw`, and your AI agent has a persistent, structured, collaboratively-trained knowledge base it can query accurately — across sessions, across departments, across months of KB drift.

Fully multimodal. Drop in `.docx`, `.pdf`, `.xlsx`, `.pptx`, markdown, images, audio, or video — lcwiki converts everything to markdown, extracts per-doc structure, concepts with cross-department family aliases, and an interactive knowledge graph in one shot. Then it lets your AI query the KB with a three-layer token-first fallback: scan 100-token tldrs → fall back to article body → only touch raw content as a last resort.

📖 The positioning

lcwiki sits at the intersection of three traditions that rarely talk to each other:

1. Personal Knowledge Management (PKM) theory — Niklas Luhmann's Zettelkasten, Tiago Forte's Building a Second Brain, the Obsidian / Logseq / Roam community: atomic notes, bidirectional links,