This glossary defines key terms used throughout the agent‑ai‑lab architecture.
It provides consistent terminology across all modules, ensuring clarity for contributors, reviewers, and implementers.
A modular, multi‑layer AI system capable of reasoning, planning, interacting, retrieving knowledge, using tools, and adapting over time.
A major architectural domain (e.g., Interaction, Cognitive & Planning, Memory & Knowledge) with clear responsibilities and boundaries.
A focused component within a layer (e.g., Retrieval Model, Planning Engine, Tool Selector).
A documented architectural unit within docs/architecture/, typically consisting of 4–5 Markdown files.
Structured cognitive processes used to interpret intent, generate explanations, solve problems, and make decisions.
Decomposing a user request into actionable steps, tool calls, or reasoning sequences.
The agent’s ability to adapt communication, infer user perspectives, and maintain socially appropriate behavior.
Modeling what the user knows, wants, or assumes, without inferring sensitive or private attributes.
Short‑term, session‑scoped memory capturing recent interactions.
Persistent storage of stable, non‑sensitive information (e.g., user preferences, system knowledge).
Embedding‑based memory enabling semantic retrieval.
Fetching relevant information from memory or external knowledge sources.
An external capability the agent can invoke (e.g., search, calculator, database query).
Choosing the appropriate tool based on intent, context, and safety constraints.
A structured sequence of tool calls and reasoning steps.
An alternative tool or strategy used when the primary one fails.
The core execution cycle that processes user input, reasoning, tool use, and response generation.
Tracking conversation state, execution state, and intermediate results.
Coordinating concurrent or sequential tasks within the runtime.
A mechanism that prevents unsafe content, actions, or tool calls.
Oversight mechanisms ensuring compliance with policies, ethics, and system rules.
Redirecting a request or event to a higher‑level safety or governance process.
A rule defining allowed and disallowed behaviors across layers.
A description of the runtime environment’s capabilities, constraints, and limits.
The system’s ability to operate consistently under varying conditions.
Resource constraints (e.g., latency, memory) that influence planning and execution.
Systematic measurement of quality, safety, and correctness.
A degradation in performance, safety, or correctness compared to previous versions.
Controlled, safe self‑optimization based on system‑wide observations.
A reversible, policy‑aligned suggestion for improving reasoning, memory, tool use, or performance.
A structured record of reasoning, tool calls, or execution steps.
A quantitative measurement used for evaluation or monitoring.
A record of decisions, safety events, or governance actions.
Assessment of how a change affects system behavior, safety, or performance.
Interaction between multiple agents with shared or complementary capabilities.
Integration of the agent with physical or simulated environments.
A controlled environment for testing agent behavior under repeatable conditions.
The first file in a module, describing purpose, responsibilities, and integration points.
Defines workflows, algorithms, or conceptual models.
Describes safety, ethics, and governance integration.
Defines logs, metrics, and auditing pipelines.