AI Agent Orchestration Platform  ·  View 13 of 32  ·  3 · Data

Memory Subsystem

Six kinds of memory, where each lives, how it is retrieved, who may read it and how long it survives.

Editable source SVG draw.io All views
Written by Store Retrieval Permission Retention Execution state Orchestrator Cosmos DB By execution id Run owner only 90 days Conversation Agent Executor Cosmos DB Recent turns Session subject Tenant policy Episodic Memory Service Cosmos + vectors Similarity + recency Agent scope grant Decay then expire Knowledge Ingestion Pipeline Azure AI Search Hybrid + reranker Security trimming Source lifecycle Profile and org Admin and user Azure SQL Direct lookup Consent flagged Erasure on request Artifacts Tool Executor ADLS Gen2 By artifact id SAS · 15 minutes Tier then delete Memory Subsystem — Scope, Store, Retrieval, Permission, Retention Application we own Data store Interface / broker Security / platform Person or role Every read is filtered by tenant and by the agent's declared memory scope, and every read is written to the audit ledger. v 1.0 · owner Data Architecture · date 2026-08

Decisions

  • Execution state and persistent memory are separate concepts with separate stores; conflating them is how a platform leaks one tenant's context into another's run
  • Every memory read is filtered by tenant and by the agent's declared memory scope, and every read is written to the audit ledger
  • Retrieval is hybrid keyword plus vector with a reranker — vector-only retrieval fails on exact identifiers, which is most enterprise queries

Retention and erasure

  • Episodic memory decays on a relevance and recency score before it expires, so the store does not grow without bound
  • Profile data is consent-flagged and supports erasure on request, propagating to the index and the artifact store
  • Retention is a policy attribute, evaluated at write time and re-evaluated by a nightly sweep

Risks

  • Security trimming on the knowledge index depends on source-system ACLs being ingested accurately; a stale ACL is an over-disclosure
  • Episodic memory can encode a mistaken conclusion and then reinforce it; memory writes from unverified agent output are opt-in per agent
◀  Data Architecture and Storage Zones All views Context Management  ▶
Solution Architecture v1.0 · Data & AI Global Practice · 2026-08 · 32 views · Microsoft Azure13-memory-subsystem