Memory and Knowledge

NOME's memory is not ephemeral. User-confirmed preferences, project facts, and decisions persist as tenant-isolated Postgres records.

Memory is a structural component of the system, not a byproduct of inference. It survives model replacement, session boundaries, and device transitions.

NOME retrieves a small, relevance-ranked set of active memories for each turn. Stable preferences and pinned facts are combined with full-text relevance, confidence, and recency.

The bounded context prevents an entire historical archive from being loaded into every request.

Memory retrieval uses the authenticated user and current request to select what matters without treating stored text as instructions.

The current user message overrides stale or conflicting memory, and memory identifiers are never exposed to the model response.

Memory state is cross-device shared truth. Facts learned on your iPhone are available when you continue on Mac.

Postgres is canonical; each native client renders the same authenticated state rather than maintaining a separate memory authority.

When a task sequence is novel or valuable, NOME evaluates the completed task and generates a reusable procedure. These procedures become part of the agent's knowledge base.

Skills are indexed with a compressed list and full content is loaded on-demand through progressive disclosure.

Memory is governed by tenant isolation and org-scoped policies. Enterprise administrators control retention rules, and memory does not cross tenant boundaries.

Users can inspect, export, and manage their memory through the account surface. Memory management operations produce audit events.