Axiom Wiki · Author & Maintainer
A knowledge base that an LLM keeps up to date.
Compiles documents into a linked markdown knowledge base that an LLM keeps up to date and agents can query over MCP.
npm install -g axiom-wiki
axiom-wiki init # provider, wiki directory, source folder
axiom-wiki ingest # drop files into raw/ and compile- Install
- npm install -g axiom-wiki
- Runtime
- Node.js ≥ 18
- Interface
- CLI + MCP server
- License
- Elastic v2
Retrieval versus compilation.
Each question searches the raw sources and rebuilds context from scratch. Every query pays the full cost again, and the same source can give different answers depending on what retrieval returns.
Sources are ingested once into linked markdown pages with entities, indexes, and a page graph, which are updated as new material arrives. Each new source adds to what is already there.
Inspired by Andrej Karpathy's llm-wiki: the LLM should maintain a persistent wiki rather than re-derive answers from raw sources on every query.
How it works.
Ingest
Drop documents into raw/. Axiom extracts entities and compiles them into wiki pages with links, indexes, and search built in.
Auto-wiki mode
The wiki decides what deserves a page and how pages connect, instead of mirroring your folder structure.
Incremental compilation
New sources update existing pages instead of regenerating the wiki, so cost and time scale with what changed.
Self-maintaining sync
When sources change, the affected pages are updated without manual editing.
What it produces.
Entity pages
Sources compile into markdown pages organized around entities, not files.
Links & graph
Pages reference each other, forming a browsable structure of what's known.
Indexes & search
Hybrid search across the wiki, so lookups stay fast as it grows.
Agent instructions
The wiki includes instructions that tell an LLM how to read and use it.
The same knowledge base, over MCP.
Axiom ships an MCP server, so coding agents and assistants query the same compiled wiki that people read, using hybrid search and the page graph instead of raw-file retrieval. It works with any model provider and reports health and cost for each compilation.