# LLM Wiki

> Your agent keeps a markdown wiki for you in Obsidian: it files each source once, cross-links the pages and flags contradictions.

LLM Wiki is a Claude Code plugin from Alireza Rezvani's claude-skills collection, a community project inspired by Andrej Karpathy's LLM Wiki pattern. Instead of digging through your documents again for every question, your agent reads each source once and files it into a persistent, interlinked markdown vault you can open in Obsidian: source summaries, entity and concept pages, comparisons and a running synthesis. You choose the sources and ask the questions; the agent does the filing and the upkeep.

What you can do with it
- Start a vault for a topic with /wiki-init: folders, an index, a log and the schema file your agent reads.
- Ingest a paper, article or note with /wiki-ingest: the agent talks through the takeaways, writes a summary page, then updates the related entity and concept pages, the index and the log.
- Ask questions with /wiki-query: it reads the index first, answers with citations to wiki pages and offers to file a good answer back as a new page.
- Run a health check with /wiki-lint: contradictions, stale claims, orphan pages, missing cross-references and concepts that deserve their own page.
- See what was added and when with /wiki-log.

What's inside
- A skill that teaches the agent the vault's layout, its ingest, query and lint workflows, and its rules.
- Subagents for each job: wiki-ingestor, wiki-librarian and wiki-linter.
- Python scripts that need only the standard library: set up a vault, rebuild the index, append to the log, BM25 search over the wiki, lint it, analyze its link graph and export a page as a Marp slide deck.
- Vault templates (CLAUDE.md, AGENTS.md, page templates for entities, concepts, sources, comparisons and syntheses) and a small example vault.

How to install
In Claude Code, run /plugin marketplace add alirezarezvani/claude-skills, then /plugin install llm-wiki@claude-code-skills. Each vault also gets an AGENTS.md, so Codex CLI, Cursor, Gemini CLI and more can work in the same vault.

Good to know
- Your sources go in raw/ and are never edited; everything the agent writes goes into wiki/.
- It's plain markdown on your own machine, with no account or server. Obsidian is recommended for reading the vault and its graph view.
- It suits knowledge you keep adding to over weeks (research, a book you're reading, team notes), not one-off questions about a single document.
- It's a community project, not made by Obsidian or Anthropic.

- Type: Plugin
- Made by: alirezarezvani
- Category: Productivity
- Formats: Claude plugins
- Version: 2.14.1
- Website: https://github.com/alirezarezvani/claude-skills/tree/main/engineering/llm-wiki
- Source: https://github.com/alirezarezvani/claude-skills/tree/HEAD/engineering/llm-wiki
- Page: https://pluginboard.ai/plugins/llm-wiki

## Works with

- Claude
- GitHub Copilot
- Claude Code
- OpenClaw

## Install

### Claude

In Claude (web or desktop): Customize → Plugins → Add → Add marketplace, enter alirezarezvani/claude-skills, then add llm-wiki

### Claude Code

In Claude Code:

```text
/plugin marketplace add alirezarezvani/claude-skills
/plugin install llm-wiki@claude-code-skills
```

---
From Pluginboard (https://pluginboard.ai).