AIAO
Agentic AI Ontology. Outlined boxes are classification dimensions; filled boxes are classes, roles and capabilities. Arrows are properties: an agent plays roles and has capabilities. Activity is kept apart for what an agent does and has no classes yet.
AgentO
OWL ontology for agentic AI workflows: agents, tasks, goals, capabilities, teams and workflow patterns. Nodes are classes; arrows are object properties drawn from domain to range. Drag nodes, click one to read it.
The project page also lists LLMAgent, HumanAgent, Prompt and Config as classes. They are not in the version 1.1 file, so they are not drawn here. Role is a free-text data property (agentRole) on Agent, not a class.
MBAD: model-based agent definition
An agent ontology anchored in the four RFLP views of model-based systems engineering. Each lane is one view; each pair is a row of the paper's Table 1, an RFLP concept linked to the agent concept it defines.
Agent types in MBAD
The paper distinguishes agents by how they reason: deterministic rule-based symbolic agents, probabilistic data-driven neural agents (for example LLM planners or RL policies), and neuro-symbolic agents that combine both. Execution phases are Sense, Decide and Act. Role appears once, as "Persona", with no class.
Crosswalk
Nearest equivalents across the four vocabularies. These are ESL's proposed alignments for discussion, not mappings published by the other authors.
| Concept | AIAO v2.2 | AgentO 1.1 | MBAD | KRT 0.1 |
|---|
Classifying agents by function: AIAO has a Functional Role dimension; AgentO records role as free text; MBAD records function as Skill and classifies agents by reasoning paradigm; KRT records the scope granted to an agent as a Mandate.