Welcome to your LLM-Wiki

This note is auto-generated by the Karpathy LLM Wiki plugin on first run. You don't need to edit it — read once, then start ingesting.

How to verify the install

This note is in English because the LLM is not yet configured. To enable wiki generation:

Open Settings → Karpathy LLM Wiki → LLM Provider.
Pick a provider, enter the API key, select a model.
Click Test Connection — it must show ✅ before proceeding.
Click Save Settings, then run Karpathy LLM Wiki: Recreate Wiki Welcome Note from the command palette.

How to use this plugin

Open the command palette with Ctrl/Cmd + P and search "Karpathy LLM Wiki". The first command below is the only one you need on day one.

Command	What it does
`Karpathy LLM Wiki: Ingest multiple files`	Pick N source notes; the plugin extracts entities / concepts / sources from each and writes wiki pages. Day one — start here.
`Karpathy LLM Wiki: Ingest single source`	Same as above, for one file.
`Karpathy LLM Wiki: Ingest from folder`	Ingests every file in a chosen folder (e.g. `inbox/2024/`).
`Karpathy LLM Wiki: Query Wiki`	Opens the right-side chat panel for asking questions against the ingested content.
`Karpathy LLM Wiki: Lint wiki`	Runs the Lint pipeline (dead links, orphans, duplicates). Use when the wiki has ~30+ pages.
`Karpathy LLM Wiki: View Ingestion History`	Opens a panel listing what each previous Ingest call created/updated.
`Karpathy LLM Wiki: Recreate Wiki Welcome Note`	Re-creates this note in the current wiki language.

The right-side Query panel can also be opened by the chat-bubble ribbon icon.

What the wiki structure means

After you ingest a source note, the plugin writes a small set of pages into your wiki/ folder. Knowing the three core types — and the optional Schema layer on top — is the single most useful piece of background for the day you start curating the wiki by hand.

The three core page types

entities/ — Named things: people, organizations, projects, products, events, places. A single source note typically produces several entity pages. Each entity page holds aliases, a summary, the source notes it appears in (mentions_in_source), and links to related entities and concepts.
concepts/ — Topics, methods, definitions, fields of study, recurring themes. "PPR", "cardiology", "schema-driven design" are all concepts. Concept pages link to other concept pages; entity pages link to concept pages.
sources/ — One page per ingested source note, with the original content plus a source_file frontmatter field. Source pages are the provenance anchor — every entity / concept page lists the source pages that mention it, so a reader can drill from a topic back to the originating note.

The Schema layer (optional)

You can enable Schema from Settings → Karpathy LLM Wiki → Schema. When enabled, the plugin maintains a wiki/schema/ folder that encodes your wiki's vocabulary — the controlled list of tag categories, section templates, and entity/concept types.

The schema lives in a single Obsidian-format page: wiki/schema/config. Open it to see the active vocabulary; the plugin rewrites this page whenever the vocabulary changes.

Ingest prompts are bound to this vocabulary, so the LLM picks tags and section headings from a fixed list rather than inventing free-form text.
When the LLM notices drift (e.g. a new concept not in the vocabulary), the suggestion appears in the Lint report modal — you accept / reject before it lands.
When the vocabulary changes, every existing page is rewritten to match (auto-backup to .llm-wiki-backups/schema/, rotation MAX_BACKUPS=3).

Without Schema, the plugin still works — the three core types are always created. Schema adds structure that makes queries more reliable on a mature wiki.

The wikilink graph

Every [[wiki-link]] between two pages is a relationship the LLM established at ingest time. The plugin uses this graph (not embeddings) for query retrieval — see the v1.23.0 release notes for the graph engine architecture. Practical takeaway: a well-curated wikilink graph is the wiki's "search index". You can add or edit [[X]] links by hand in any page, and the next query will pick them up.

The folder layout

All wiki files live under wiki/ (configurable in Settings → Wiki folder):

wiki/
├── entities/    # Named things (people, orgs, projects)
├── concepts/    # Topics, methods, definitions
├── sources/     # One page per ingested note (provenance)
├── schema/      # Optional vocabulary + section templates
├── index.md     # Auto-generated graph index
└── log.md       # Auto-generated activity log

Quick start

Pick a few source notes to ingest. Run Karpathy LLM Wiki: Ingest multiple files from the command palette. Tick the notes you want, then click Add to queue. The modal stays open so you can watch progress.
Wait for ingest to finish. Each note takes 10–60 seconds (LLM extraction). You can keep working — completions appear as Notices. View Ingestion History lists every batch's result.
Try a query. Open the right-side Query Wiki panel (chat-bubble ribbon icon) and ask something about your content.
Tune settings if needed. Settings → Karpathy LLM Wiki: language, wiki folder, schema, tag vocabulary, auto-watch. Defaults are sensible for new vaults.

Full guide in the README: github.com/GD4AI/obsidian-llm-wiki