Independent Memory in current agentic systems is often fragmented across model- provider features, application databases, session histories, framework-specific stores, unstructured files, and retrieval indexes. This document distinguishes temporary context supplied to an inference request from persistent memory that remains addressable, machine-readable, and governed beyond a single request. It specifies a provider-independent architecture and data model for persistent memory in agentic systems based on Memory Scope isolation, typed and versioned Memory Objects, machine-readable provenance, append-only Event Ledger history, separate lifecycle, availability, retention, and validation state, and isolated Derived Indexes and Embedding Spaces. The architecture separates a transient Compute Plane from an authoritative Persistent State Plane and treats embeddings, lexical indexes, graph projections, generated retrieval summaries, ranking caches, and retrieval caches as non-authoritative derived state. The architecture is independent of model provider, storage engine, and protocol transport while allowing future bindings and multiple independent implementations. It does not claim that related systems or standards do not exist; rather, it defines a common architectural vocabulary and interoperability target for persistent agentic memory.