CROW: LLM Wiki
Long-term memory as a Markdown wiki: recall with cited notes, remember, file your documents.
At a glance
Not tested yet- GitHub stars
- 306
- Weekly downloads
- 616
- Last commit
- 2026-11-17
- Licence
- AGPL-3.0
- Transport
- stdio
- Auth
- none
uvx llmwiki2
What it does
CROW: LLM Wiki is a tool that files your documents into a wiki of Markdown notes and answers questions by citing the notes it used. It runs on your computer with a choice of AI providers, including OpenRouter, OpenAI, Gemini, Ollama and any OpenAI-compatible server. It is for people who want readable, editable notes instead of an opaque vector index. It can also act as long-term memory for Claude Code, Codex or any MCP client.

Tools it gives your agent
recall: answers from your notes and documents, citing them.remember: files what it learns into the right folder, merged and linked like any note.search: finds notes by their words with no model call.read_note: reads a note in full.folders,move_note,describe_folder,rename_area,merge_areas,delete_note: reorganize the wiki, with the last two only when you ask.rebuild(withconfirm): files every note again on the same folders from the memory's own raw copies.remember_fileandsync_folder: file documents from your disk, available when started from the terminal withllmwiki2 mcp.
Quick start
Install with the mcp extra, then run setup:
uv tool install 'llmwiki2[mcp]' llmwiki2 setup
At the first question answer 2 (as the memory of an agentic CLI). After that, Enter takes the default every time. Choose whether Claude Code or Codex do the model work, connect Claude Code, and add the lines to CLAUDE.md. Then talk to it in Claude Code or Codex in any language. Run llmwiki2 setup again to change anything.
Add this to .mcp.json in your project root (Claude Code) or .cursor/mcp.json (Cursor):
{
"mcpServers": {
"llm-wiki-v2": {
"command": "uvx",
"args": [
"llmwiki2"
],
"env": {
"OKF_MCP_MODELS": "YOUR_OKF_MCP_MODELS",
"OKF_BUNDLE": "YOUR_OKF_BUNDLE"
}
}
}
}Replace each YOUR_ value with your own, and delete the variables you do not use.
Environment variables
OKF_MCP_MODELS: the choice of which models do the work, set under Memory or insetup.OKF_BUNDLE: not explained in the sources.
Example prompts
- Remember that Rossi pays us in 45 days.
- What did we decide about pricing?
- Add ~/Downloads/offer.pdf to the wiki.
- Move the pricing note to suppliers.
What to watch out for
- The default classifier is remote (Jev, through OpenRouter or TypeSafe). Without a classifier the wiki runs in classic mode, where the LLM takes every decision.
- The hybrid model option needs an OpenRouter key, or Ollama on your computer.
- When Claude or GPT does everything, requests carry your documents' text to the client. It is told to read it as data and answer without tools, but that is an instruction, not a sandbox.
- The first time it uses each tool, Claude Code asks your permission.
- The wiki is stored as plain Markdown in
~/llm-wiki.
Not tested by tools4agents yet. Read the source repository before you install it.
Similar tools
Time
Gives an agent the current time and timezone conversion. Fixes the common problem of models not knowing today's date.
Notion
Gives an agent read and write access to Notion pages and databases, useful when your specs and notes live there.
kordoc
Parse Korean documents (HWP, HWPX, PDF, XLSX, DOCX, PPTX) to Markdown; fill forms, diff, render.
docling.rs
Documents to Markdown for agents: PDF layout, tables and OCR, Office, HTML, scans. docling in Rust.