This project provides an open-source implementation of Karpathy's LLM Wiki, designed to automate the creation and maintenance of personal knowledge bases. It addresses the common challenge of knowledge base upkeep, which often becomes burdensome for humans. By leveraging an LLM (Claude via MCP), users can upload documents, and the system automatically generates, cross-references, and maintains wiki pages, allowing knowledge to compound over time. This benefits researchers, engineers, and power users seeking an efficient, LLM-assisted knowledge management solution.

How It Works

The system employs a three-layer architecture: Raw Sources (immutable user documents), The Wiki (LLM-generated markdown pages), and The Tools (LLM interaction layer). A Next.js frontend interfaces with a FastAPI backend, which utilizes Supabase (Postgres) for data storage and an S3-compatible service for file uploads. Core to its operation is an MCP (Model Communication Protocol) server that enables Claude.ai to directly interact with the knowledge vault. Claude.ai uses provided tools to search, read, write, and delete content, automating tasks like summarizing new sources, updating entity pages, flagging contradictions, and maintaining cross-references, thereby reducing the manual maintenance burden to near zero.

Quick Start & Requirements

The fastest way to try LLM Wiki is via the hosted service at llmwiki.app. Sign up, upload sources, and connect your Claude.ai account using the provided MCP configuration.

For self-hosting, prerequisites include:

• Python 3.11+
• Node.js 20+
• A Supabase project (or local Docker setup)
• An S3-compatible bucket
• A Claude.ai account

The setup involves initializing the database, running the API and MCP servers, and starting the web frontend, requiring configuration of environment variables for database credentials, API keys (optional for OCR), and storage.

Highlighted Details

• Automated wiki generation, cross-referencing, and maintenance by an LLM.
• MCP server allows direct LLM interaction with the knowledge vault via tools for search, read, write, and delete operations.
• Document ingestion automatically updates multiple wiki pages and flags knowledge inconsistencies.
• Querying synthesized knowledge can create new wiki pages, enabling knowledge compounding.
• Includes a linting feature for knowledge base health checks.
• Supports PDF, articles, notes, and office documents (via conversion); PDF OCR is available via Mistral API.

Maintenance & Community

The provided README does not contain specific details regarding notable contributors, sponsorships, or community channels like Discord or Slack.

Licensing & Compatibility

The project is licensed under the Apache 2.0 license. This permissive license is generally compatible with commercial use and linking within closed-source applications.

Limitations & Caveats

The README does not explicitly detail limitations such as alpha status or known bugs. Self-hosting requires managing multiple external services (Supabase, S3) and an external LLM provider (Claude.ai). OCR functionality is dependent on obtaining and configuring a Mistral API key.