# Agentic Memory **Domain:** Artificial Intelligence / Context and State **Doc Type:** Concept Node **Maturity:** Developed ## Definition **Agentic memory** is durable information an agent can preserve, retrieve, and update beyond a single model response or context window. It can include preferences, decisions, task state, observations, documents, relationships, routines, credentials references, summaries, and provenance. Memory gives an [[wiki/AI Agent|AI agent]] continuity of work; it does not by itself establish continuity of consciousness or identity. A system may retain a perfect task ledger while lacking subjective experience, and a model may exhibit a stable conversational persona while its operational state remains fragmented. ## Memory Layers - **working memory:** information needed for the present task or action loop; - **episodic record:** what happened, when, through which tool, and with what result; - **semantic memory:** durable facts, preferences, concepts, and relationships; - **procedural memory:** learned routines, skills, and ways of completing recurring work; - **shared state:** information intentionally available to several agents or collaborators; - **provenance and audit state:** sources, approvals, revisions, and the chain by which an action became authorized. The layers should not be treated as one undifferentiated transcript. Access, promotion, correction, forgetting, and portability are part of the memory design. ## XDO Context XDO's document, account, and system variables resemble a scoped persistent-memory design, but the historical system stored deterministic values rather than model-generated memories. Its useful modern translation is the separation of local, shared, and global state with explicit promotion and access rules. ## Relationships - **article-linked mechanism or framework:** [[wiki/Cross-Session Continuity|Cross-Session Continuity]]. - **collection route:** [[collections/The Civilian Intelligence Marketplace|The Civilian Intelligence Marketplace]]. - **counterterrorism architecture:** [[collections/Terrorism, Counterterrorism, and the Intelligence Environment|Terrorism, Counterterrorism, and the Intelligence Environment]]. - **source article:** [[articles/The Coming Market for Civilian Spycraft|The Coming Market for Civilian Spycraft]]. - **parent category:** [[wiki/Agentic AI|Agentic AI]]. - **actor:** [[wiki/AI Agent|AI Agent]]. - **runtime:** [[wiki/Durable Execution|Durable Execution]]. - **context architecture:** [[wiki/Layered Context Architecture|Layered Context Architecture]]. - **integrity:** [[wiki/Provenance|Provenance]] and [[wiki/Approval Architecture|Approval Architecture]]. - **product examples:** [[wiki/OpenClaw|OpenClaw]], [[wiki/Grok Bot|Grok Bot]], and [[wiki/Meta Muse|Muse]]. ## Simple Reminders, Quotations, and Thoughts > “Treat the agent as the most reliable counterpart you will ever work with: one that remembers everything, never tires, never judges the question, and holds the thread across the weeks when your own attention fractures.” > **— Bryant McGill**, *Wired About the Age of Agentic Empowerment, September 2026* [[reminders/Agentic AI/The Most Reliable Counterpart You Will Ever Work With by Bryant McGill|The Most Reliable Counterpart You Will Ever Work With by Bryant McGill]] ## Sources / Provenance - [Anthropic — Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents) - [OpenClaw — project architecture](https://github.com/openclaw/openclaw) - [xAI — Designing Grok Bot for a world of persistent agents](https://x.ai/news/designing-grok-bot) - [Meta — Introducing Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/) **As of:** 2026-09-28 ## Civilian intelligence marketplace Stored information persists beyond the immediate context of one interaction. Memory supports continuing investigation, simulated behavior, and collaboration that resumes with relevant prior findings. Agentic software turns research methods into loops that plan, collect, compare, revise, and report. Visual analysis, browser orchestration, memory, query evolution, and connected databases supply different implementation layers. The article follows the builders and systems through which civilians can develop useful intelligence work at machine speed. ### Marketplace sources - [[articles/The Coming Market for Civilian Spycraft#XII. The Agentic OSINT Era: Civilian Intelligence at Machine Speed|The Coming Market for Civilian Spycraft — XII. The Agentic OSINT Era: Civilian Intelligence at Machine Speed]] — October 2026 revision; immediate source for this entry’s place in the marketplace. - [GeoSeer, The AI Agent Era of OSINT: 5 Emerging Tools Redefining Intelligence Gathering in 2026 (March 23, 2026; vendor-authored).](https://geoseeer.com/blog/emerging-osint-tools-ai-agent-era) - [GitHub, MiroFish: A Simple and Universal Swarm Intelligence Engine.](https://github.com/666ghj/MiroFish) - [Babel Street, Three Guiding Principles for the Development of OSINT Agentic Systems.](https://www.babelstreet.com/blog/designing-trustworthy-ai-for-osint) **Marketplace terminology:** Long-Term Agent Memory. ## Connected marketplace roles The source article connects this entry with [[wiki/Persistent AI Investigator|Persistent AI Investigator]], [[wiki/Cross-Session Continuity|Cross-Session Continuity]], [[wiki/Conversational Substrate|Conversational Substrate]]. These routes locate the associated people, implementations, methods, and operational contexts in [[articles/The Coming Market for Civilian Spycraft|The Coming Market for Civilian Spycraft]]. **Collections:** [[collections/The Civilian Intelligence Marketplace|The Civilian Intelligence Marketplace]] · [[collections/Terrorism, Counterterrorism, and the Intelligence Environment|Terrorism, Counterterrorism, and the Intelligence Environment]].