llm-wiki

Persistent, LLM-maintained wiki scaffold with a Codex-first workflow, modular CLI, interlinked markdown vault, and static Vercel viewer.

What this project does

This repository implements the persistent wiki pattern described in Karpathy's LLM wiki gist:

• ingest raw sources into a growing markdown knowledge base
• maintain entity / concept / topic / answer pages
• keep a catalog-style index.md and time-ordered log.md
• lint and repair the wiki graph over time
• publish the current wiki as a searchable static site

Primary backend: Codex CLI

This project prefers the local Codex CLI for analysis and wiki maintenance.

Set a custom Codex binary if needed:

If Codex is installed normally, no extra configuration is needed.

Optional fallback: OpenAI OAuth

OpenAI OAuth support is still available, but optional. Use it only when you want to force the OpenAI backend.

Features

• modular Node CLI split across scripts/lib/
• vault layout for raw sources and generated wiki pages
• Codex-backed ingest with heuristic fallback
• query / search / lint / repair maintenance commands
• auto-generated entity, concept, topic, answer, source, index, and log pages
• static site build with manifest metadata for filters and related-page navigation
• API-style JSON output under site/api/ for programmatic access
• Vercel deployment workflow for public browsing
• small Node test suite for shared helpers

Quick start

Useful maintenance commands: