Concepts

Understand memory models, architecture, and design tradeoffs. Use how-to guides for procedures and reference for complete option lists.

If you are new to the project, read the first three pages below in order: the three memory types, then why the graph, then the POLE+O model. That gives you the vocabulary the rest of the pages assume.

Memory concepts

Understanding the three memory types — Short-term, long-term, and reasoning memory, and why the library keeps them separate.
Why Neo4j? Graph memory architecture — Why a graph, rather than a vector store or relational database, backs agent memory here.
The POLE+O data model — The Person/Object/Location/Event/Organization classification long-term memory uses for entities.
Bolt and NAMS backends — How the self-hosted Bolt backend and the hosted NAMS backend compare and where they diverge.
How entity extraction works — How the spaCy, GLiNER2.5, and LLM extraction stages combine into one pipeline.
Entity resolution and deduplication — How entity resolution decides two mentions are the same entity, and when it flags instead of merges.
Structured extraction — How structured output from an LLM is validated and turned into entities and relationships.
Why the provider protocol? — Why LLM and embedding providers are a swappable protocol rather than hard-coded integrations.
Choosing an agent framework integration — How the framework integrations compare, to help you pick one for your agent.
Cross-agent memory sharing — What happens, and what can go wrong, when multiple agents share the same memory graph.
Why the v0.2 operational primitives are opt-in — Why multi-tenancy, buffered writes, and consolidation ship opt-in rather than as defaults.
Understanding the three-tier context model — How recent messages compress into observations, and observations into reflections, as context grows.
How the NAMS AI provider retrieves memory — Why the separate NAMS AI provider package merges five memory sources by taking turns instead of ranking them by confidence.
Ontologies — What a NAMS ontology is and how it constrains extraction for a workspace.
Agent Skills (Preview) — What a NAMS Agent Skill is and how it differs from an ontology.