my-llm-wiki

Drop any files into a folder. Get a living, queryable knowledge graph.

The idea

In April 2026, Andrej Karpathy shared a concept he called LLM Wiki — a personal knowledge system with three layers:

Raw — your files, never modified
Compile — structured knowledge with cross-references
Query — ask questions without re-reading source files

The key insight: compile once, query forever. The wiki is a persistent, compounding artifact — it grows with every session.

my-llm-wiki implements all three layers. One command builds the graph. The living wiki cycle keeps it growing.

pip install my-llm-wiki
cd your-project && llm-wiki .

The living wiki cycle

Monitor → Rebuild → Lint → Write-back → Report
   ↑                                       │
   └───────────────────────────────────────┘

Each session adds knowledge. Insights get filed back. The graph compounds over time — exactly what Karpathy envisioned.

What you get

wiki-out/
  graph.html       ← interactive graph (vis.js)
  graph.json       ← persistent graph data
  WIKI_REPORT.md   ← god nodes, surprising connections
  wiki/            ← Wikipedia-style articles
  vault/           ← markdown vault with [[wikilinks]]
  cache/           ← SHA256 cache (skip unchanged files)

See How It Works for the full pipeline architecture.

v0.9 — Semantic vault maintenance & session capture

Per-heading page citations: PDFs, Word docs, and slides now track which page each heading came from (visible in vault YAML and CLI output).

Session capture: llm-wiki capture scans your Claude Code logs and extracts note candidates. Review them, promote to vault with llm-wiki note.

Vault maintenance: /wiki maintain agent runs semantic audits — detects orphans, broken links, contradictions, stale TODOs.

New query helpers: llm-wiki query orphans and llm-wiki query stale-refs power the audit workflow.

CI matrix: Tests on Linux, macOS, Windows × Python 3.10–3.13. Integration tests on Ubuntu + Python 3.13.