SamurAIGPT/llm-wiki-agent

A personal knowledge base that builds and maintains itself. Drop in sources — Claude (or Codex/Gemini) reads them, extracts knowledge, and maintains a persistent interlinked wiki. Works with Claude Code, Codex, OpenCode, Gemini CLI. No API key needed.

What it solves

LLM Wiki Agent transforms a collection of disparate documents (PDFs, DOCX, etc.) into a structured, interlinked markdown wiki. Unlike traditional RAG systems that retrieve raw chunks of text during a query, this tool compiles knowledge into persistent pages that accumulate and compound over time, automatically creating cross-references and flagging contradictions between sources.

How it works

The system operates as a skill for coding agents (such as Claude Code, Codex, or Gemini CLI). It uses a set of schema files (like CLAUDE.md) to instruct the agent on how to maintain the wiki.

When a user triggers an ingest command, the agent reads the source documents—converting non-markdown files via markitdown—and extracts knowledge to create or update entity pages (people, companies), concept pages (ideas, frameworks), and a living overview. It also builds a knowledge graph using NetworkX and Louvain community detection, generating a self-contained graph.html file that visualizes both explicit wikilinks and inferred semantic relationships.

Who it’s for

  • Researchers who need to synthesize information from multiple academic papers and reports.
  • Students or Book Readers who want to build a thematic map of a book's characters and arguments.
  • Knowledge Workers managing personal knowledge bases or business intelligence (meeting transcripts, project docs).
  • Competitive Analysts tracking market trends and company announcements over time.

Highlights

  • Persistent Knowledge: Creates structured markdown pages that persist across sessions rather than relying on ephemeral chat history.
  • Automated Synthesis: Automatically generates entity and concept pages and updates a living overview of all ingested material.
  • ** uma Contradiction Detection**: Flags conflicting claims between new and existing sources at the time of ingestion.
  • Multi-Format Support: Ingests a wide variety of formats including PDF, DOCX, PPTX, XLSX, and HTML.
  • Obsidian Integration: Designed to be compatible with Obsidian, allowing users to browse their agent-generated wiki as a vault.
  • Visual Knowledge Graph: Produces an interactive HTML visualization of the knowledge network.