Agent Memory
The persistence layer through which an agent carries state across invocations — prior decisions, retrieved documents, learned preferences — distinct from both the model weights and the context window.
- StatusIndustry term
- AdoptionEmerging
- LayerIntelligence
Why it matters
Memory is the component that governance frameworks have not noticed. Weights are versioned and reviewed; memory accumulates silently and changes behavior with no model change at all. Two deployments of an identical model with different memory states are different systems, and no current model risk process would catch it.
Where it comes from
Convergent usage across agent frameworks from 2024 onward.
No published instrument defines this term yet; the note above is its origin.
What it is not
These are routinely confused with Agent Memory. The distinctions have consequences for how a system is governed.
- Context Window
Memory is what gets selected INTO the window. The window is the budget; memory is the store.
- RAG corpus
A retrieval corpus is shared and curated. Memory is per-agent and accrues from the agent's own history.
How it connects
What it needs
- Agent Memory depends on Agent Memory Concept
- Agent Memory requires Audit Trail Canon term
MARQUE questions it addresses
- AuditabilityCan a decision be reconstructed months later, exactly?
Not to be confused with
- Context Window Canon term
Related framework objectives
Connected through the MARQUE (a reference framework organized around eight questions) crosswalk, an editorial mapping. These objectives from the Financial Services AI Risk Management Framework are assigned to the same questions that this term addresses. The ones sharing most of this term's vocabulary are shown first, as a reading aid. The assignment is to the question, not to the term: the source does not map this term to any objective directly.
Question: Auditability · 131 control objectives assigned
- MP-3.5.4Oversight Effectiveness Evaluation and ImprovementMap
- GV-1.2.1AI Principles and Policy IntegrationGovern
- GV-1.4.2AI Risk Documentation StandardsGovern
Cite and reuse
Cite the identifier QIS-TERM-00010, not the URL: identifiers are stable, URLs may change. This entry is free to read, quote and index under the dual license. Corrections sent to editor@qisfund.com are published.
Technical details: identifier, graph relationships and record
| Predicate | Target identifier | Target | Predicate meaning |
|---|---|---|---|
dependsOn | concept:agent-memory | Agent Memory | The subject cannot function correctly without the object. |
requires | term:audit-trail | Audit Trail | The subject imposes the object as a precondition. |
addresses | marque:auditability | Auditability | The subject speaks to the object as a question or concern. |
notEquivalentTo | term:context-window | Context Window | The two are routinely conflated and are distinct. |
- Canonical identifier
QIS-TERM-00010- Tier
- Canonical industry term
- Domain · layer
- Context & memory · Intelligence
- Semantic aliases
- None recorded
- First published
- 2026-08-02
- Last reviewed
- 2026-08-02 · 180-day cycle