Concepts

Explanation-oriented pages. Read these before the guides if you want to understand why luplo behaves the way it does.

• Philosophy — the five refusals (vectors-don’t-lead, five-hop, decisions-immutable, typed-and-bounded edges, not-a-general-memory) and the three operational commitments that enforce them.

• Positioning — how luplo differs from generic AI-memory tools on eight axes, and when it is (or isn’t) the right tool.

• Architecture — the three interfaces (CLI / MCP / HTTP) sharing one core, the Backend protocol, and the worker.

• Data model — the twelve tables, items as substrate, and the item_types registry.

• Search pipeline — the four-stage retrieval pipeline, strict-first glossary, and the role of vector reranking.