llm-wiki
base698
A reusable starter kit for running a markdown knowledge base as an LLM-maintained wiki — the pattern from Andrej Karpathy's "LLM as Wiki" gist.
Open base698/llm-wiki → Checked against the four rules.
Works with any agent
ships AGENTS.md, the cross-tool instruction file.
In its own words
A reusable starter kit for running a markdown knowledge base as an LLM-maintained wiki — the pattern from Andrej Karpathy's "LLM as Wiki" gist. Instead of RAG (rediscovering knowledge from raw documents on every question), the LLM incrementally builds and maintains a persistent, interlinked wiki between you and your sources. Each source is read once and integrated everywhere it matters; the synthesis compounds.
Division of labor: the human curates sources, explores, and asks questions; the LLM does all the bookkeeping — summarizing, cross-referencing, filing, indexing — which is exactly the maintenance burden that makes humans abandon wikis.
Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase. — Karpathy
Bootstrap a new wiki
- Copy this directory (minus this README's kit framing) to wherever the wiki lives,
git init, commit.
- Edit the
Folder mapsection ofCLAUDE.md/AGENTS.md— replace the placeholder
topic folders with your actual domains (e.g. Claims_ML/, Platform/, Vendors/).
- Drop your first document into
Sources/and tell your agent: "ingest this."
It will write the wiki page(s), update index.md and Concept Map.md, and append to log.md.
- From then on: ask questions (the agent answers from the wiki, with citations, and
files durable answers back), drop sources, and occasionally ask for a lint.
Quoting 36% of the README. Read the rest on GitHub ↗
Read from the repository's git tree on 2026-08-17.