llmwiki plugin
Turns session history of Claude Code, Codex, Cursor and Gemini CLI into a searchable wiki on Karpathy's LLM Wiki pattern, locally
- 400 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The tool reads the full session histories of all your agents in the home directory, and they can hold keys, passwords and personal data
- The installer can add a SessionStart hook to ~/.claude/settings.json, and the built site can be published to GitHub Pages
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/llmwikiDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The tool reads the full session histories of all your agents in the home directory, and they can hold keys, passwords and personal data
- The installer can add a SessionStart hook to ~/.claude/settings.json, and the built site can be published to GitHub Pages
This is third-party code. Review the repository files before installing.
What it does
llmwiki collects the transcripts of your agent sessions into a personal knowledge base. Adapters read the .jsonl files of Claude Code, Codex CLI, Copilot, Cursor and Gemini CLI, plus an Obsidian vault. The sessions become Markdown files with sensitive data redacted, kept in the raw/sessions folder. The `llmwiki build` command turns it into a static site with Cmd+K search, an activity heatmap, code highlighting and a dark theme. Your agent writes the wiki on Andrej Karpathy's LLM Wiki pattern. The /wiki-ingest command turns raw sessions into `sources`, `entities`, `concepts`, `syntheses`, `comparisons` and `questions` pages linked with double-bracket wikilinks. Sixteen lint rules find broken links, orphan pages, duplicates and stale summaries. Pages carry a confidence score and a lifecycle from draft to archived. For other AIs the site serves llms.txt and llms-full.txt files, a JSON-LD graph and plain-text twins of every page. An MCP server with twelve tools starts with `python3 -m llmwiki.mcp`, and a SessionStart hook can sync new sessions on every Claude Code launch.
Who it is for. For developers who have piled up hundreds of agent sessions over months and want to search them for solutions and link them into a knowledge base.
Good fit when
- You need to find how you solved a task in past Claude Code or Codex sessions without paging through transcripts by hand
- You want a personal wiki on the Karpathy pattern: source, entity and concept pages with cross-links
- You work with several agents and want the history of all of them in one site with agent labels
- Your notes live in Obsidian: the tool connects to the vault and brings Dataview queries and Templater templates
- You need an export for other AIs: llms.txt, llms-full.txt and a JSON-LD graph
Not a fit when
- Sessions held passwords, keys or client data that must not be passed to an agent for processing
- You need a fast command line to search history without a wiki: separate tools exist for that
- You need a shared cloud service for a team: llmwiki runs locally and keeps data in project folders
Example request
Run /wiki-ingest for the new sessions and show which concept pages appeared about database migrationsLimitations
Python 3.9 or newer is required. Dependencies are minimal: the standard library and the `markdown` package, with no database and no npm. Code highlighting loads from a CDN at view time, so it does not work offline.
The `llmwiki sync` command only fills raw. To get meaningful pages in wiki/sources you need either an agent with the /wiki-ingest command or a synthesis backend in examples/sessions_config.json: `ollama` for a local Ollama or `dummy` for a stub.
The site is served at 127.0.0.1:8765 and can be published to GitHub Pages. The raw folder is in .gitignore; making sure session data does not reach a public output is up to you.
How to disable. If you accepted the SessionStart hook during setup, remove its entry from ~/.claude/settings.json. Then delete the llm-wiki repository folder together with raw, wiki and site, which hold your data. The tool does not change the original agent session files.
FAQ
How do I install llmwiki?
Clone the repository, then run ./setup.sh on macOS and Linux or setup.bat on Windows. The installer creates the raw, wiki and site folders, installs the Python package, detects your agents and enables the matching adapters. It can also offer the SessionStart hook and run the first sync at once. The pip route is pip install -e .
Which agents and sources does llmwiki read?
Adapters cover Claude Code, Codex CLI, Copilot, Cursor, Gemini CLI and an Obsidian vault. The `llmwiki adapters` command shows for each whether it is available on the machine and configured. You can limit a sync to one source with the `--adapter` flag, for example `--adapter claude_code`.
Do I need API keys or a server?
No. Everything runs locally. Your agent writes the wiki pages with the /wiki-ingest command. For automatic synthesis in the CLI you can point synthesis.backend at a local Ollama.
How are the data layers arranged in llmwiki?
The first layer on the Karpathy pattern is immutable sessions as Markdown in raw/sessions. The second is the wiki pages the model writes: `sources`, `entities`, `concepts`, `syntheses`, `comparisons`, `questions`, plus `index.md`, `overview.md` and `log.md`. The build adds the site folder with a static site and the AI exports.
How do I connect llmwiki to Obsidian?
The `link-obsidian` command symlinks the project into a vault, after which Obsidian graph view, backlinks and search work with the wiki. A Dataview dashboard with ten queries and four Templater templates come with it. The `--vault PATH` flag writes output straight into an Obsidian or Logseq vault.
Discussion
No questions or reviews yet.
Related
PluginsLLM Wiki
llm-wiki
Knowledge bases that AI compiles: topics, import, llm-wiki lint checks, the /wiki:ingest and /wiki:archive commands. A plugin for Claude Code and Codex
A workflow for Claude Code, Codex and Gemini CLI where the agent files source documents into a persistent, linked wiki
A skill set that lets an agent run a personal Obsidian wiki: it compiles knowledge from documents and past sessions and answers with citations
LLM Wiki Compiler
llm-wiki-compiler
A knowledge compiler: turns raw sources into an interlinked wiki with answer verification and approval, with an MCP mode
A plugin and skill set that turns an Obsidian vault into a knowledge base with sources, links and grounded answers
Karpathy LLM Wiki
karpathy-llm-wiki
A skill for a personal knowledge base the model maintains: ingest sources, compile cited articles, lint integrity
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| llmwiki (this entry) | no | 400 | |
| Karpathy LLM Wiki | no | 2,624 | a skill for a personal knowledge base the model maintains, without a site or session collection |
| LLM Wiki Compiler | no | 2,223 | a CLI compiler that turns raw sources into an interlinked wiki with answer verification |
| Coding Agent Session Search (cass) | no | 1,196 | a CLI and TUI for fast search across agent session history, without a wiki or pages |
README in short
The project shows a live demo built from synthetic sessions, screenshots of the home, session list and session pages, and a 90-second tutorial on the commands from `llmwiki init` to `llmwiki all`. Separate sections cover automation, quality control, Obsidian integration, project maintenance and end-to-end browser tests. The CLI command reference closes the document.