A forkable template for building LLM-maintained personal knowledge bases using GitHub Copilot.

Instead of starting from scratch on every query, Copilot incrementally builds and maintains a persistent, interlinked wiki of markdown files that compounds with every source ingested.

This project is an implementation of the LLM-Wiki concept popularized by Andrej Karpathy, designed to create a "Compounding Knowledge Pattern" through purely file-based local storage.

The system is divided into three distinct layers of responsibility:

1. Raw Sources (raw/): Your collection of source documents. These are immutable—the AI reads them but never modifies them.
2. The Wiki (wiki/): Generating interlinked markdown files. The AI owns this layer—it writes it; you read it.
3. The Schema (.github/copilot-instructions.md): The "brain" or configuration layer. It tells the AI how to be a disciplined wiki maintainer rather than a generic chatbot.

1. Fork this repository: Click the "Fork" button to create your own copy.
2. Clone and Open in VS Code: Open the repo in an environment with GitHub Copilot installed.

For rapid setup, you can skip the UI and use the terminal:

• Scaffold only (fastest): npx degit SriSatyaLokesh/copilot-llm-wiki#main my-wiki
• Fork and Clone (official): gh repo create my-wiki --template="SriSatyaLokesh/copilot-llm-wiki" --public --clone

1. Enable Prompt Files:
   • Open VS Code Settings (Ctrl+,).
   • Search for github.copilot.chat.promptFiles.
   • Set it to true.
2. Drop your first source: Put a markdown file or a URL reference in the raw/ directory.
3. Run Ingest:
   • In VS Code Chat: Type /ingest (or use the paperclip icon to attach .github/prompts/ingest.prompt.md) and provide- Batch Mode: Drop multiple markdown files in raw/ and run:

• Bash: ./.github/skills/wiki-ingest/scripts/intake.sh
• PowerShell: .\.github\skills\wiki-ingest\scripts\intake.ps1 ).

Tell Copilot to "ingest ". It will:

• Fetch and save the source to raw/.
• Extract key facts.
• Create/update pages in wiki/entities/ and wiki/concepts/.
• Update the wiki/index.md and wiki/log.md.

Ask a question about your knowledge domain. Copilot will:

• Read wiki/index.md to find relevant pages.
• Consult the interlinked wiki pages.
• Answer with citations.
• Offer to file the answer as a new QA page in wiki/qa/.

Run "lint the wiki" via lint.prompt.md or the librarian agent to check for orphans, broken links, or contradictions.

1. Open .github/copilot-instructions.md.
2. Find the [YOUR DOMAIN] placeholders and replace them with your specific domain (e.g., "Medical Research", "Codebase Documentation", "Legal Case Files").
3. Customize the wiki/overview.md to reflect your project's goals.

If you have the GitHub Copilot CLI installed, you can invoke the dedicated librarian agent:

copilot --agent librarian -p "ingest raw/new-data.md"

The librarian agent automatically decides between direct ingestion and using the intake scripts based on the complexity of your request.

To get the most out of your LLM Wiki, we highly recommend using Obsidian to view your wiki/ directory.

• Knowledge Graph: Use Obsidian's "Graph View" to see how your entities and concepts are interconnected.
• Easy Navigation: Clickable links, back-links, and local previews make exploring your knowledge base feel like a "second brain."
• Markdown-Native: Since the wiki is 100% standard markdown, Obsidian handles it natively without any conversion needed.

Inspired by Andrej Karpathy's llm-wiki. Built by the community.