Storage keeps objects. Memory keeps meaning in motion.

A folder can preserve a file. A database can preserve a row. Neither automatically preserves how a decision was reached, which source challenged it, or why a concept later changed.

For AI agents, that missing context is consequential. The agent needs more than an archive; it needs a living model that can be inspected and improved.

A memory lifecycle needs boundaries

Formation rules decide what becomes a candidate memory.
Consolidation rules decide when two observations belong together.
Recall rules decide what context is relevant to the present task.
Revision rules decide how a human correction propagates without rewriting history.

Each boundary should be visible. Automatic processing may propose structure, but people must be able to name, merge, split, and reject it.

The ontology is the working surface

In Consilience, teams edit the ontology as the shared surface for concepts and relationships an agent should understand. EngramRAG preserves that structure through time, and agents work from the current version.

That is the practical meaning of an Ontology OS: a shared, revisable world that both people and agents can work in.