Framework 07 of 12 · Cross-session Phase · Longitudinal intelligence
Agent Memory & Context Continuity Framework
The design of what the agent knows about you over time
The design of what the agent knows about you over time - how that knowledge is encoded, persists across sessions and channels, how users can inspect and correct it, and how it is governed under data protection regulation. The personalisation and continuity layer.
Agent Memory &: Core Principles
Memory Is a Design Material
What the agent remembers about the user is not a technical implementation detail - it is a design decision with profound implications for the experience. Memory enables personalisation, continuity, and efficiency. But it also creates privacy risks, bias accumulation, and dependency. The framework treats memory as a first-class design material that requires intentional architecture.
Users Must Be Able to Inspect and Edit Memory
The agent's memory about a user must be inspectable and editable by that user. If the agent believes the user prefers organic produce, the user should be able to see that belief, correct it if wrong, and delete it if unwanted. Memory opacity is a trust violation.
Consent Governs Memory Formation
The agent should not silently accumulate knowledge about the user. Memory formation requires consent - either explicit (the user agrees to preference tracking) or contextual (the user's actions in a specific context imply consent for that context). The framework defines consent architectures for different memory types.
Memory Has a Taxonomy
Not all memory is equal. Ephemeral memory (this session only), session memory (this interaction), persistent memory (across sessions), and cross-channel memory (across platforms) each have different governance requirements, consent thresholds, and design implications.
Forgetting Is a Feature
The ability to forget - to delete memory, reset preferences, or start fresh - is as important as the ability to remember. Users must have clear, accessible pathways to erase agent memory, either selectively or completely. In GDPR jurisdictions, this is a legal requirement. In all jurisdictions, it is a trust requirement.
The agent that remembers everything about you is not a good agent. The agent that remembers the right things, forgets when asked, and lets you see what it knows - that is a trustworthy agent.
Agent Memory &: Implementation Patterns
Memory Taxonomy
A classification system for agent memory types: ephemeral (discarded after use), session (persists within an interaction), persistent (survives across sessions), and cross-channel (shared across platforms). Each type has defined governance rules, consent requirements, and retention policies.
When to use: When designing the memory architecture for any agentic system.
Memory Inspection and Editing Interfaces
User-facing interfaces that display what the agent knows about the user and allow correction, deletion, or modification. Includes preference dashboards, belief inventories, and memory timelines that show when and how each memory was formed.
When to use: As a required feature in every agentic system that maintains user memory.
Consent Architecture for Memory Formation
Multi-layered consent systems that govern what the agent may remember: opt-in for sensitive categories, contextual consent for routine preferences, and explicit consent for cross-channel memory sharing. Includes consent withdrawal patterns that trigger appropriate memory deletion.
When to use: During onboarding and at every point where the agent forms new persistent memory.
GDPR-Aligned Data Minimisation
Patterns for collecting and retaining only the minimum memory necessary for the agent's function. Includes purpose limitation (memory tied to specific use cases), storage limitation (automatic expiry for unused memory), and data minimisation (storing abstractions rather than raw data where possible).
When to use: As a foundational constraint in all memory architecture decisions, especially in EU/UK deployments.
Cross-Session Context Continuity
Patterns for maintaining meaningful context across sessions without requiring the user to re-explain their situation. Includes session handoff summaries, preference persistence, and progressive context building that enriches the agent's understanding over time.
When to use: At every session boundary where the user returns to an ongoing relationship with the agent.
Memory Reset and Erasure Pathways
Clear, accessible mechanisms for users to delete agent memory - selectively (remove specific beliefs or preferences) or completely (full memory wipe). Includes confirmation patterns, consequence previews (what will change if this memory is deleted), and verification that deletion is complete.
When to use: As a required feature accessible from any point in the user experience.
Agent Memory &: Commerce Applications
Purchase History and Preferences
In agentic commerce, the agent's memory of past purchases, brand preferences, size information, and quality expectations enables increasingly efficient delegation. The framework governs how this commercial memory is formed, maintained, and made available to the user for inspection and correction.
Cross-Retailer Preference Portability
When a consumer uses agents across multiple retailers, their preferences should be portable - not locked into a single platform. The framework defines standards for preference export, import, and synchronisation that give consumers control over their commercial memory.
Seasonal and Contextual Adaptation
The agent should remember that the user's preferences change with context - different food preferences in summer versus winter, different gift preferences for different recipients, different budget constraints at different times of the month. The framework provides patterns for contextual memory that adapts without requiring explicit re-specification.
Preference Decay and Refresh
Consumer preferences change over time. The agent must detect when stored preferences are stale and prompt for refresh rather than acting on outdated information. The framework defines decay models for different preference types and refresh patterns that feel helpful rather than intrusive.
Memory without consent is surveillance. Memory without inspection is opacity. Memory without deletion is imprisonment. The Memory & Context Continuity Framework ensures that agent memory serves the user, not the system.
Agent Memory &: Guidance for Teams
Start With
- -Classify every piece of user data your agent stores by memory type
- -Build a memory inspection interface showing what the agent knows about each user
- -Implement selective and complete memory deletion pathways
- -Define consent requirements for each memory type in your system
Build Toward
- -Federated memory architectures that keep data under user control
- -Cross-platform memory portability standards
- -Automated preference decay detection and refresh prompts
- -Privacy-preserving memory that stores abstractions rather than raw data
Measure By
- -Memory accuracy - how often do stored preferences match current user intent?
- -Inspection utilisation - how often do users review their agent memory?
- -Deletion completion rate - are deletion requests fully honoured?
- -Consent coverage - what percentage of stored memory has explicit consent?
Agent Memory &: Lifecycle Connections
Framework 01
Intent Architecture Framework
Memory enriches intent specification. An agent with good memory can help users specify intent more efficiently by referencing past preferences.
Explore frameworkFramework 12
Ethical Constraint & Value Alignment
Memory governance is an ethical design challenge. Value Alignment Architecture defines the moral boundaries of what agents should remember.
Explore frameworkFramework 09
Explainability & Observability
Memory inspection is a form of observability. Users need to understand not just what the agent knows, but how it uses that knowledge in decisions.
Explore frameworkAgent Memory &: What Comes Next
Memory & Context Continuity governs what the agent knows over time. The next framework - Absent-State Audit - evaluates what the agent did while the human was away.
Agent Memory &: The Framework Ecosystem
Navigate the complete lifecycle of Agentic Experience Design. Each framework addresses a distinct phase of the human-agent relationship.