Lore Graph

A self-hosted knowledge graph + hybrid retrieval pipeline for D&D / Forgotten Realms lore, exposed to a DM companion over MCP. It models people, organizations, goals, and reified events across in-world time, scoped to where and when a party is, and can generate plausible instigators for an event you want to seed.

See CLAUDE.md for architecture decisions and PLAN.md for the phased build plan.

Quickstart

cp .env.example .env          # set NEO4J_PASSWORD, ANTHROPIC_API_KEY
make up                       # start Neo4j (browser :7474, bolt :7687)
make schema                   # apply constraints + seed
pip install -e .              # install the package
make test                     # run the validator test suite
make extract-demo             # network-free demo of the validation layer

Status

Schema and the extraction + validation layer are in place and tested. The LLM call is isolated behind lore_graph.extraction.extract_with_llm. Next up: the idempotent loader, gazetteer bootstrap, sourcebook parser, and the orchestrating pipeline — then ingest Rise of Tiamat end to end. See PLAN Phases 1–3.

Layout

• schema/ — Cypher schema (constraints, conventions, seed, flagship queries).
• lore_graph/ — the pipeline package. extraction.py is built; loader.py, parsing.py, pipeline.py are stubs with documented contracts.
• data/sources/ — raw books, gitignored. Never commit copyrighted text.