LLM Wiki L1 L2 plugin
A Claude Code plugin that runs a Karpathy-style wiki on Logseq or Obsidian and splits knowledge into memory rules and a deep wiki
- 169 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The plugin writes wiki pages and commits into your graph or vault, and ingested sources land in files in full
- The prune --l1 command changes and deletes Claude Code memory files after confirmation
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/llm-wiki-l1-l2Detects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin writes wiki pages and commits into your graph or vault, and ingested sources land in files in full
- The prune --l1 command changes and deletes Claude Code memory files after confirmation
This is third-party code. Review the repository files before installing.
What it does
llm-wiki implements Andrej Karpathy idea of a model-maintained wiki. The plugin installs into Claude Code and asks for one value, the wiki root: your Logseq graph or Obsidian vault folder. Then `/llm-wiki setup` asks which tool you use and which namespaces you want and creates the schema, dashboard, hub pages, access log and an `llm-wiki.yml` file. Commands: `/llm-wiki ingest <source>` processes a link, file or text and updates 5 to 15 pages with cross-references, `/llm-wiki query` searches in two stages through hub indexes and answers with source attribution, and `/llm-wiki lint [--fix]` checks the wiki against 13 rules: orphan pages, stale content, broken references, index drift and credential leaks. `/llm-wiki prune` evicts cold pages from the live index and `/llm-wiki status` shows metrics. The architecture has two layers, like a CPU cache. L1, Claude Code memory of 10 to 20 files, loads in every session and holds rules, gotchas and personal preferences. L2, a wiki of 50 to 200 pages, is read on demand and holds projects, workflows and research. Everything is written to local disk and changes are recorded as local git commits.
Who it is for. People who keep a personal knowledge base in Logseq or Obsidian and want Claude Code to file facts from links, files and work notes into it.
Good fit when
- After a release the outcome is logged with `/llm-wiki ingest "deployed v2.0 to production"` and updated pages come back
- A question to accumulated notes goes through `/llm-wiki query` and returns an answer with sources
- Once a week `/llm-wiki lint` runs to find orphans, stale pages and leaked keys
- Rarely needed project knowledge goes into the wiki while rules the agent cannot do without stay in Claude Code memory
Not a fit when
- You have neither Logseq nor Obsidian: the plugin is built around their page format and schema
- Several Claude sessions write to the wiki at once: parallel edits can conflict
- You have a ten-page note set: a schema of five page types and eight lint rules will feel like overkill
Example request
Run /llm-wiki ingest for this article, update the related pages and show what changedLimitations
Claude Code 2.1.275 or newer is needed for the one-line install, and the author tested it on 2.1.296. The plugin asks for the wiki root, and any sources you ingest end up in wiki files, including names and contact details if the source has them. With `memory_path` set, the skill reads local Claude Code memory, and `prune --l1` changes, moves or deletes its files after your confirmation. The network is used only by `ingest` to fetch a page from a link. Parallel sessions writing to the wiki can conflict, and the schema starts paying off at around 50 pages.
How to disable. Run `/plugin uninstall llm-wiki`. If you installed a copy, delete `.claude/commands/wiki.md`. The wiki and log stay in your Logseq graph or Obsidian vault and can remain as ordinary notes.
FAQ
How do I install llm-wiki in Claude Code?
Run `claude plugin marketplace add MehmetGoekce/llm-wiki` and `claude plugin install llm-wiki@mehmetgoekce`. On Claude Code 2.1.275 and later one line in a session is enough: `/plugin install llm-wiki --marketplace MehmetGoekce/llm-wiki`.
What are L1 and L2?
L1 is Claude Code memory, loaded in every session and holding rules and gotchas. L2 is the Logseq or Obsidian wiki, read on demand through `/llm-wiki query`. The routing rule: if a mistake without the knowledge would be dangerous or embarrassing, it goes to L1.
Where does the plugin send data?
Nowhere. According to the author, the project has no backend or telemetry, and pages and the log live in your graph or vault. Only `ingest` uses the network, to fetch a page from a link.
What does lint check?
Thirteen rules: pages with no incoming links, stale content, missing properties, broken references, index drift, credential leaks and due L1 claims. The `--fix` flag repairs some findings.
Does it work without the plugin?
Yes. Copy `SKILL.md` to the project `.claude/commands/wiki.md`, replace `${user_config.wiki_path}` and `${CLAUDE_PLUGIN_ROOT}` with two paths and call commands as `/wiki`. The copy stays at the chosen version.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code plugin for Obsidian: stores sources, compiles linked wiki pages and answers from the accumulated knowledge.
Turns session history of Claude Code, Codex, Cursor and Gemini CLI into a searchable wiki on Karpathy's LLM Wiki pattern, locally
LLM Wiki Manager keeps a personal Markdown wiki through an agent: it ingests sources, links pages and answers with citations
An Obsidian vault template where an agent follows a CLAUDE.md schema to compile sources into a linked wiki and extend it.
Claude Code plugin for a personal wiki per Karpathy's pattern: the agent grows pages from sources and answers with citations.
A CLI that creates an Obsidian-ready Markdown vault and installs four skills so an agent maintains the knowledge base.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| LLM Wiki L1 L2 (this card) | no | 169 | |
| LLM Wiki | no | 1,528 | a plugin with /wiki:ingest and /wiki:archive commands and a lint check, while this has two memory layers, L1 in Claude Code and L2 in Logseq or Obsidian |
| Karpathy LLM Wiki | no | 2,732 | a skill for a personal knowledge base with sources, cited articles and integrity checks, while this adds two memory layers and a Logseq or Obsidian wiki |
| claude-obsidian | no | 16,397 | a plugin for an Obsidian vault, while this also supports Logseq and adds a rules layer in Claude Code memory |
README in short
The README begins with a two-layer diagram and the Karpathy idea, followed by a quick start with and without the plugin, the L1 and L2 architecture and a command table with the ingest flow. The page schema, a Logseq and Obsidian comparison, before and after examples, configuration, trade-offs and a privacy section come next.