No description
| argo-workflows | ||
| argocd | ||
| backups | ||
| calcom | ||
| cert-manager | ||
| ci | ||
| drone | ||
| fenec | ||
| fenics | ||
| formbricks | ||
| gitea | ||
| grafana | ||
| harbor | ||
| hermes-agent | ||
| hermes-bridge | ||
| hermes-webui | ||
| homelab-root | ||
| kubeopencode | ||
| kubermetrics | ||
| manim | ||
| mcp | ||
| medusa | ||
| minio | ||
| model-runner | ||
| n8n | ||
| notebooks | ||
| nvidia | ||
| odoo | ||
| ollama | ||
| shopilab | ||
| skopeo | ||
| sops-secrets-operator | ||
| supabase | ||
| trash | ||
| zellij | ||
| .gitignore | ||
| .gitkeep | ||
| _template.md | ||
| README.md | ||
atoms/
Knowledge atoms, organized by topic-branch. Source of truth.
Layout
atoms/
├── README.md ← you are here
├── _template.md ← copy this when creating a new atom
│
├── <branch-1>/ ← one folder per topic-branch
│ ├── 2026-01-15-some-claim.md
│ ├── 2026-02-03-another-claim.md
│ ├── _archive/ ← superseded atoms live here
│ └── ...
│
├── <branch-2>/
│ └── ...
│
└── ...
One folder per topic-branch. One file per atom. One claim per file.
See _template.md for the frontmatter and body format. See CLAUDE.md at the repo root for the full spec.
Why atoms exist
Karpathy's original LLM Wiki goes raw → wiki directly. The atom layer is this repo's main addition — it solves three problems:
- Loss of information. Wiki compilation is lossy. Without atoms, you can't recover what got dropped without re-reading raw.
- False sense of source of truth. Wiki looks authoritative, but it's a derived artifact. Atoms are the truth; wiki is a cache.
- Provenance. Every wiki claim should be traceable to atoms, and every atom traceable to raw. Without an atom layer, the chain breaks at the wiki.
When a wiki page is wrong, you fix the underlying atom and recompile. You never patch the wiki directly.
What's gitignored
Everything in this folder except README.md, _template.md, and .gitkeep is gitignored. Your atoms are personal content; the framework is what gets shared.