Skip to content

johbrust/llm-wiki-agent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LLM-Wiki-Agent

This project defines an agent to create an LLM-Wiki as proposed by Andrey Karpathy. This implementation is a simplyfied version of the much more comprehensive Obsidian Wiki.

Setup

The Wiki is based on Markdown files and therefore can be read in any text editor. For the full experience we recommend to use Obsidian for navigating the Wiki.

The Wiki is tool-agnostic, i.e., it can be used with OpenCode, Claude, or any other agentic tool (currently, OpenCode and Claude are supported by the setup script). A local pdf_mcp MCP server is configured for the agent to be able to read PDFs. The local MCP requires uv to run and will be started automatically. To enable OCR from images, install Tesseract with sudo apt install tesseract-ocr.

For using the agent, create a directory for the Wiki and clone this repo into that directory or add it as a submodule if the Wiki is a git repo itself.

# Local Wiki setup:
mkdir ~/awesome-wiki
cd awesome-wiki
git clone <repo-url> .llm-wiki-agent

# As a submodule:
git submodule add <repo-url> .llm-wiki-agent
git submodule init
git submodule update

To install the Wiki Agent and all necessary skills run:

cd ~/awesome-wiki

# Setup for OpenCode:
./.llm-wiki-agent/setup.sh opencode

# Setup for Claude:
./.llm-wiki-agent/setup.sh claude

Git LFS

Sources (like PDFs) are kept in the Wiki. When using git for working and sharing, it is recommended to use Git LFS (see the installation instructions if you haven't set it up already) to avoid indexing and diffing of large files.

Usage

Adding Sources

Add your source (PDF, .md, image, ...) to the input directory of the Wiki and task the agent to ingest the source with the wiki-ingest skill. Don't forget to select the Llm-Wiki agent!

User: I added a new source to the `input` directory for you to ingest.

Asking Questions

One core idea behind the LLM Wiki concept is the ability to ask questions about the content of the Wiki in natural language. The agent then reads the Wiki and can respond by also linking to the sources.

User: Explain the distinction between <concept 1> and <concept 2> in the context given by the Wiki.

Acknowledgements

This agent was developed to help with literature review note taking during my time at DFKI. The DFKI Niedersachsen (DFKI NI) is sponsored by the Ministry of Science and Culture of Lower Saxony and the VolkswagenStiftung.

About

No description, website, or topics provided.

Resources

License

Stars

1 star

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages