MemoryGrid

Core Memory

Semantic Memory

Component	Description
Rules	Formal constraints or behavioral norms agents follow, enabling policy enforcement and predictable coordination.
Knowledge / Facts	Declarative knowledge about world states, agent capabilities, environment; supports inference and shared logic.
Concepts	Abstract semantic structures or domain-specific ontologies shared between agents for symbolic interoperability.

Episodic Memory

Stores specific, temporally indexed experiences (who, what, when, where). Used for causal reasoning, trust formation, and learning across agent interactions.

Procedural Memory

Encodes learned skills or process, predefined action sequences, procedures, protocols, and routines for task execution. Essential agent behavior.

Working Memory

Provides temporary storage o short-term context store for current goals, active tasks, local perceptions, recent interactions, or transient computations allowing agents to manipulate and combine data during active problem-solving and decision-making.

Reflections Memory

Captures meta-cognition: summaries & self-assessments about past decisions, strategy adaptations, mistakes, and introspective patterns used to guide future self-modification or strategy evolution.

World Model

Component	Description
Environment Models	Represent spatial, temporal, and causal dynamics of the agent's operational context, enabling situated action.
Small-Scale Models of Reality	Internal simulation engines used to run predictive models, test strategies, and evaluate potential future states.
Social World Models	Encodes associations, norms, roles, protocols, and collective behaviors within multi-agent societies to support coordination.
Self Models	Encapsulate an agent’s own identity, norms, capabilities, history, and belief states for self-awareness and regulation.
Agency Models	Represent the agent’s own structure of autonomy, drive, objectives, culture, values, norms, regulation machinery.
Reward Models	Define value functions or heuristics guiding agent decision-making, aligned with system or mission-level goals.

Communication Memory

Stores, organizes, indexes historical interaction logs between agents or with humans. Enables coherent dialogue, recall of prior intents or commitments, and supports long-term relationship modeling across agents in the system.

Reward Memory

Captures episodic records of received rewards, both intrinsic and extrinsic. This helps agents evaluate action effectiveness over time, adapt policies based on accumulated value trends, and develop reward-based strategy shifts.

Context Cache

A short-term memory buffer holding transient situational information such as recent observations, immediate task context, or local state. Enables timely decisions, context-sensitive responses, and rapid adaptation to ongoing environmental changes.

Temporal & Forms

Temporal

Component	Description
Short Term Memory	Temporarily holds recent signals, decisions, and interactions. Crucial for reactive behaviors, immediate planning, and working memory operations in agents.
Long Term Memory	Retains durable knowledge, policies, and learned experiences. Serves as the foundation for historical reasoning, value alignment, and longitudinal adaptation.

Forms

Component	Description
Vector Memory	Encodes memory as high-dimensional embeddings. Enables similarity-based retrieval and fast neural access for perception, classification, and semantic queries.
Tree Memory	Organizes memory hierarchically (e.g., decision trees, plan structures). Supports branching logic, recursive reasoning, and modular knowledge representation.
Graph Memory	Represents relationships as nodes and edges. Useful for storing semantic networks, social interactions, knowledge ontologies, and causal dependencies among agents or concepts.

Strategic Memory

Plans Memory

Goals & Task Memory

Maintains both long-term goals and short-lived task objectives, supporting context-aware prioritization, back tracking, task conflict handling, and dynamic goal switching.

State Memory

Records internal and external state traces relevant to strategic reasoning such as belief states, or environmental situation snapshots.

Normative Memory

Encodes rules, commitments, and institutional norms & policies agents must respect or enforce. This underpins social contract adherence and protocol-abiding behavior.

Value Memory

Component	Description
Learned Preferences	Emergent patterns of choice shaped by experiential feedback.
Ethical Principles	Prescribed moral constraints that influence decision boundaries.
Culture	Encoded social norms, roles, and symbolic knowledge for shared meaning.
Utility Function	Computable evaluative mechanism for expected outcome ranking.

Reward Memory

Captures accumulated reward histories and patterns that shape learning signals, behavioral tuning, and reinforcement alignment over time.

Shared Memory

Shared Knowledge Base

A unified repository of common facts, concepts, models, or protocols that all agents can read from and contribute to. It ensures consistent situational awareness and semantic grounding.

Consensus Memories

Distributed memory units that store group-agreed facts, collectively endorsed goals, or resolved positions. Useful for democratic reasoning, collaborative planning, and collective belief formation.

Global Workspace

A broadcast-capable shared interface where prioritized memories, goals, or signals are elevated for temporary global attention across agents. Inspired by Global Workspace Theory (GWT).

Blackboard System

A modular coordination framework where agents post partial results, intermediate inferences, or collaboration requests to a shared “blackboard.” Other agents asynchronously read and contribute.

Memory Processes

Memory Ingestion

Memory Acquisition

Process by which raw inputs, observations, or interactions are collected and preprocessed to be stored as memory traces.

Memory Encoding

Transformation of acquired inputs into structured memory representations, making them compatible with internal formats like vectors or graphs.

Memory Ops

Memory Inference

Deriving new insights or facts from existing memories through reasoning, prediction, or symbolic mechanisms.

Indexing

Organizing memory entries with keys, metadata like tags, or relational structures for efficient lookup and semantic access.

Matching

Comparing current inputs or queries against stored memories to identify patterns, similarities, or previously encountered states.

Search & Retrieval

Locating and extracting relevant memories based on contextual, content-based, or symbolic criteria to support real-time cognition.

Utility Maximizer

Attention Mechanism

Prioritizes focus on the most relevant, urgent, or novel memories and signals by filtering context, noise and guiding processing.

Contextualization Engine

Interprets retrieved memories within the current task or situation, adapting them to present goals, agents, and environmental cues.

Shared Memory

Memory Registry

Process by which raw inputs, observations, or interactions are collected and preprocessed to be stored as memory traces.

Memory Routing

Process by which raw inputs, observations, or interactions are collected and preprocessed to be stored as memory traces.

Access Control

Process by which raw inputs, observations, or interactions are collected and preprocessed to be stored as memory traces.

Distributed Persistent DB

Process by which raw inputs, observations, or interactions are collected and preprocessed to be stored as memory traces.

Distributed Storage

Process by which raw inputs, observations, or interactions are collected and preprocessed to be stored as memory traces.

Distributed Cache

Process by which raw inputs, observations, or interactions are collected and preprocessed to be stored as memory traces.