Glossary
Quick reference for AiQarus terminology.
A
Agent
A configured AI entity that performs tasks autonomously. Defined by a system prompt, capabilities, constraints, and behavior rules.
Agent Definition
The immutable configuration of an agent, including its prompts, model settings, tool access, and constraints. Once activated, definitions cannot be changed.
Audit Trail
The complete, cryptographically-verified record of all events during an agent’s execution. Uses SHA-256 hash chaining for tamper evidence.
C
Capability
A permission granted to an agent to use a specific tool or access a specific resource. Agents start with zero capabilities.
Constraint
A safety limit applied to an agent, such as maximum steps, maximum duration, or token limits.
D
DAG (Directed Acyclic Graph)
The structure used for multi-agent workflows. Ensures workflows have clear dependencies and no circular references.
Decision Point
A moment during execution when an agent pauses for human approval. Created based on risk thresholds or policy rules.
E
Episodic Memory
Long-term memory that stores experiences from past runs, enabling agents to learn from history.
Evaluation
The validation of an agent’s output against a schema or semantic criteria. Results in pass/fail with details.
H
Hash Chain
A sequence of cryptographic hashes where each hash includes the previous hash. Used to create tamper-evident audit trails.
Human-in-the-Loop (HITL)
A design pattern where AI agents pause for human review and approval at key decision points.
I
Interactivity Mode
Configuration that determines how often an agent pauses for human input: autonomous, key_decisions, or conversational.
M
Memory Namespace
A scoped storage area within the memory system. Controls which agents can read/write to specific memory locations.
Multi-tenancy
Architecture where multiple organizations share the same platform while maintaining complete data isolation.
O
Organization
The top-level tenant entity in AiQarus. All data is scoped by organization.
Observe
The fourth phase of the TDAO loop. The agent summarizes results to inform the next iteration.
P
Project
A logical grouping of agents within an organization. Used for organization and access control.
R
Risk Assessment
The analysis of an agent’s planned actions to determine potential impact and required approval levels.
Run
A single execution instance of an agent with specific input. Contains multiple steps and traces.
RBAC (Role-Based Access Control)
The permission system that controls what users can do based on their assigned roles.
S
Semantic Memory
Long-term memory that stores facts, knowledge, and learned information.
Step
An individual phase execution within a run (Think, Decide, Act, or Observe).
System Prompt
The foundational instructions that define an agent’s behavior, personality, and capabilities.
T
TDAO Loop
Think-Decide-Act-Observe: The structured reasoning loop that all AiQarus agents follow.
Tool
An external capability that an agent can invoke, such as making HTTP requests, running code, or accessing memory.
Trace
A single audit event within an execution, cryptographically linked to form a tamper-evident chain.
Trigger
A mechanism that automatically starts an agent run based on time (cron), HTTP request (webhook), or system event.
W
Working Memory
Short-lived memory that exists only for the duration of a single run. Used for current task context.
Workflow
A DAG-based composition of multiple agents that execute in sequence, parallel, or conditionally.