# Memory Architecture

NEW IN 2026

 Agent memory matured from "pick a vector database" into a benchmarked production discipline in 2026, with dedicated evaluation suites (LoCoMo (Long Conversation Memory)), MemBench, MemoryAgentBench, MemoryArena) and an ecosystem spanning roughly 20+ frameworks and vector stores across managed-cloud, self-hosted, and local-MCP hosting models.

<div align="left" dir="ltr" id="bkmrk-type-storage-layer-p"><table><colgroup><col width="120"></col><col width="121"></col><col width="118"></col><col width="263"></col></colgroup><tbody><tr><td>### Type

</td><td>### Storage

</td><td>### Layer

</td><td>### Purpose

</td></tr><tr><td>Short-Term

</td><td>Session

</td><td>In-memory

</td><td>Conversation state

</td></tr><tr><td>Long-Term

</td><td>Cross-session

</td><td>Vector DB

</td><td>Semantic recall

</td></tr><tr><td>Graph Memory

</td><td>Persistent

</td><td>Graph DB

</td><td>Relationships

</td></tr><tr><td>Episodic

</td><td>Persistent

</td><td>DB

</td><td>Prior tasks / outcomes

</td></tr><tr><td>Procedural

</td><td>Persistent

</td><td>DB

</td><td>Learned workflows

</td></tr><tr><td>Policy Memory

</td><td>Persistent

</td><td>DB

</td><td>Rules &amp; restrictions

</td></tr><tr><td>Audit Memory

</td><td>Permanent

</td><td>SQL

</td><td>Logs &amp; traceability

</td></tr></tbody></table>

</div>### Three Patterns for Memory Control

● Pattern A — Context-resident: everything lives in the context window with compression. Simple, but caps out fast on long-running agents.

● Pattern B — Retrieval-augmented (workhorse pattern): working memory in-context, long-term records in a vector or structured store, injected each step. Recommended default, the engineering burden is manageable and the main challenge is retrieval quality.

● Pattern C — Tiered memory with learned control: multiple tiers (context, structured DB, vector store, cold archive) managed by a learned or prompted controller. Highest headroom, highest engineering cost — graduate to this only when data shows Pattern B is the bottleneck.

## Best Practice

Memory entries should include:

● source

● timestamp

● confidence

● owner

● retention policy

● trust score

Known open problem across 2026 memory systems: selective forgetting. Most benchmarked systems handle retrieval and test-time learning reasonably well but still fail conspicuously at deciding what to evict. Budget explicit engineering time for eviction policy, not just ingestion.