LLM Wiki Agent
A workflow for Claude Code, Codex and Gemini CLI where the agent files source documents into a persistent, linked wiki
Low risk
We rate an entry low when it mostly gives the agent instructions and reference material.
Why this level
- Only works with local markdown files inside the repository
Install
Manual install
git clone https://github.com/SamurAIGPT/llm-wiki-agent.git
cd llm-wiki-agentClone the repository locally
This is third-party code. Review the repository files before installing.
What it does
The repository defines a structure and instructions (CLAUDE.md, AGENTS.md) that the agent follows to maintain a wiki itself. The user drops documents into raw/ and asks the agent to ingest them; the agent reads the source, creates source, entity and concept pages under wiki/, and updates index.md and the append-only log.md. Separate local scripts under tools/ run structural checks without an LLM (health.py), LLM-based content quality checks (lint.py), and build a link graph (build_graph.py) into graph.json and graph.html. Non-markdown files (PDF, DOCX, PPTX, XLSX, HTML and others) are first converted via markitdown.
Who it is for. For researchers and analysts who want an agent to build a linked knowledge base out of documents they keep dropping in.
Good fit when
- You need to accumulate material from many sources into one linked wiki across sessions
- You want the agent to answer questions by synthesizing across already ingested sources
- You need to check the wiki for contradictions and orphan pages
Not a fit when
- You need a one-off translation or summary of a single document, not an accumulating base
- You need a multi-user wiki with accounts, not local markdown files
Example request
Ingest this paper into the wiki and show how it connects to sources already thereLimitations
No API key is needed for Claude Code since the agent maintains the wiki through its own file read and write tools. health.py and build_graph.py need a local Python install. lint.py makes LLM calls for semantic checks. The wiki is entirely local, with no sync between machines and no server.
How to disable. Delete the cloned repository folder or stop opening it in the agent.
Security check
- Only works with local markdown files inside the repository
README in short
The README calls the project a coding agent skill: drop sources into raw/, tell the agent to ingest them, and it extracts knowledge and builds a linked wiki with an index.md catalog, an append-only log.md, a synthesizing overview.md, and source, entity, concept and synthesis pages. Setup is a plain git clone opened in Claude Code, Codex, OpenCode or Gemini CLI, with no API key or separate Python setup required. Claude Code additionally exposes /wiki-ingest, /wiki-query, /wiki-lint, /wiki-graph slash commands. MIT licensed.
FAQ
Do you need a separate API key?
Not for Claude Code and similar agents, they work through their own file access. A key is only needed for the lint.py script, which makes LLM calls for quality checks.
Which source formats are supported?
Markdown is ingested directly, while PDF, DOCX, PPTX, XLSX, HTML, TXT, CSV, JSON, XML, RST, EPUB and other formats are first converted to markdown via markitdown.
Related
A self-hosted knowledge base with block-level references and a built-in MCP server for connecting AI agents to your notes
A CLI for every Google Workspace API with JSON output and agent skills: Drive, Gmail, Calendar, Sheets and more
Local search over Markdown notes, docs and meeting transcripts: keywords, semantic search and reranking, with an MCP server
A task manager for AI-driven development: breaks a PRD into dependent tasks and guides the agent through them via MCP or CLI