LLM Wiki Compiler

•
2.2k
•

Compiles raw documents, transcripts, and web pages into an interlinked, citation-backed Markdown wiki for human exploration and AI context.

Inspired by Andrej Karpathy's LLM Wiki pattern, LLM Wiki Compiler moves the effort of relationship extraction and context discovery from query time to compile time. It ingests diverse raw sources—such as PDFs, notes, READMEs, and URLs—and synthesizes them into structured, interlinked Markdown pages with verifiable line-level citations. Featuring runtime trust gates, customizable domain lifecycle profiles (such as research and editorial templates), evaluation metrics, and local graph visualization, it turns transient unstructured files into a durable, auditable knowledge base accessible via CLI, TypeScript SDK, or Model Context Protocol (MCP) servers.

主要功能

012,225 GitHub stars
02Two-phase compilation pipeline extracting typed concepts, entities, comparisons, and overviews with line-level citations
03Configurable Lifecycle Profiles (CLP) for defining domain schemas, directed relations, transition requirements, and review gates
04Hybrid retrieval combining semantic chunk search, BM25 reranking, and wikilink graph expansion
05Built-in local web viewer featuring graph exploration, citation chips, freshness indicators, and theme options
06Built-in MCP server and TypeScript SDK for programmatic ingestion, querying, context-pack generation, and linting

使用案例

01Building persistent, citation-backed context packs for AI agents instead of passing loose, unindexed files
02Structuring complex research or editorial workflows with validated entities, lifecycle gates, and provenance checks
03Maintaining an evolving personal or team knowledge base with automated stale-page detection and link validation