LLM Wiki

A compiled knowledge base powered by LLMs. Feed raw documents in, get structured, interlinked wiki pages out — served as a static site via GitHub Pages.

How It Works

LLM Wiki uses a compilation paradigm instead of traditional RAG:

1. Ingest — Add source documents to raw/ (manually or via GitHub Issues)
2. Compile — compile.py sends the document + existing wiki context to an LLM
3. Validate — Output is checked for valid frontmatter, source references, and wikilinks
4. Serve — MkDocs builds a searchable static site from the compiled wiki pages

All infrastructure is GitHub-native: repositories for storage, Actions for compute, Pages for hosting.

Quick Start

# Install dependencies
uv sync

# Dry run — prints compiled output without writing files
COMPILE_DRY_RUN=true \
LLM_PROVIDER=openai \
OPENAI_BASE_URL=http://localhost:8000/v1 \
OPENAI_API_KEY=local \
OPENAI_MODEL=your-model \
uv run python scripts/compile.py raw/your-document.md

# Full run — writes wiki pages and updates the index
LLM_PROVIDER=openai \
OPENAI_BASE_URL=http://localhost:8000/v1 \
OPENAI_API_KEY=local \
OPENAI_MODEL=your-model \
uv run python scripts/compile.py raw/your-document.md

# Preview the site
uv run mkdocs serve

Project Structure

llm_wiki/compile.py     # Compilation engine (parse, validate, gather, call_llm, apply)
scripts/compile.py      # CLI entry point
raw/                    # Source documents (append-only)
wiki/                   # Compiled wiki pages (LLM-owned)
mkdocs.yml              # MkDocs configuration
AGENTS.md               # System prompt loaded by compile.py

LLM Providers

Controlled by the LLM_PROVIDER environment variable:

Provider	Env Var	Use Case
openai (default)	OPENAI_API_KEY	OpenAI API or any compatible endpoint
anthropic	ANTHROPIC_API_KEY	Anthropic API

Any OpenAI-compatible endpoint (Ollama, LiteLLM, LM Studio) works by setting OPENAI_BASE_URL.

Environment Variables

Variable	Required	Description
LLM_PROVIDER	No	openai (default) or anthropic
OPENAI_API_KEY	If openai	API key (any string for local endpoints)
OPENAI_BASE_URL	No	Override endpoint (e.g. http://localhost:11434/v1)
OPENAI_MODEL	No	Model name (default: gpt-4o)
ANTHROPIC_API_KEY	If anthropic	Anthropic API key
MAX_TOKENS	No	Output token limit (default: 16384)
COMPILE_DRY_RUN	No	true = print output without writing files

Development

uv sync                # install deps
uv run pytest -v       # run tests (25 tests)
uv run ruff check .    # lint
uv run mkdocs build    # build site

License

MIT