Skip to main content

Agent Class Tree (Interactive)

This interactive diagram presents the clean, composable Agent hierarchy. Click any class to view its long definition, key facet axes (attributes), and one example instance. Use pinch/scroll to zoom and drag to pan.

Loading Agent Tree…

Tip: You can copy a node's name via the copy icon in the detail card; great for referencing in specs.

Kivanura's uniques take​

Kivanura introduces novel concepts (if you use this tree/ontology/hierarchy please give the credit to ** Kivanura foundation **)

  • ModelTriggeredAgent, StaticOrchestratorAgent, DynamicReasoningAgent, AutonomousCognitiveAgent

  • HybridAgent family (esp. AugmentedHumanAgent), HybridCollectiveAgent

  • BioComputationalAgent (organoid/fungal computing)

  • ArtificialCollectiveAgent (DAOAgent) modeled explicitly as an agent

  • Full facet system with inheritance/overrides tied to classes

Glossary: Facets & Values​

Below are concise, canonical definitions for every facet and value used in the Agent hierarchy. These are semantics, not implementation prescriptions.

1) Substrate​

  • Living — Biological tissue is the primary computational/actuating medium.
  • Artificial — Non-biological digital or electromechanical systems (software, electronics, robotics).
  • Hybrid — Mixed biological + artificial substrates in one agent.
  • Simulated (tag) — The agent exists only as a simulation (e.g., inside a virtual environment). Tag that can apply in addition to other values.
  • Quantum (tag) — Uses quantum information processing for part of cognition or control. Tag that can apply in addition to other values.

2) Embodiment​

  • SoftwareEmbodied — Entirely digital embodiment; acts through data/APIs/protocols.
  • PhysicalEmbodied — Has physical body/actuators; acts in the real world.
  • DistributedEmbodied — Realization is spread across multiple nodes/bodies/actors but exposes a unified agent boundary.

3) Intelligence Style​

  • Reactive — Maps perceptions to actions without explicit planning (reflexes, rules, PID, policies).
  • Deliberative — Uses explicit reasoning/search/constraint satisfaction; may consult policies or planners.
  • Learning — Improves policies or models from data/experience (ML/RL/online updates).
  • SelfReflective — Performs meta-reasoning about its own goals, plans, memory, or errors; can revise strategies.

4) Autonomy​

  • Assisted — Requires human approval or guidance for material actions; tightly constrained.
  • SemiAutonomous — Operates independently for bounded tasks with guardrails/hand-offs for exceptions.
  • Autonomous — Selects and executes actions end-to-end within a policy envelope and safety constraints.

5) Planning​

  • None — No explicit planning; direct/reactive execution or fixed workflow.
  • Local — Short-horizon or single-level planning (e.g., step-by-step tool selection, path segments).
  • Hierarchical — Multi-level plans (goals → subgoals → actions), e.g., HTN/MCTS with abstractions.

6) Memory​

  • Stateless — No persistence beyond a single step/transaction; decisions don’t depend on prior episodes.
  • Episodic — Short-lived per-task/per-session context (scratch state, windowed history).
  • LongTerm (Vector/Graph/Hybrid) — Durable knowledge stores (vector DBs, knowledge graphs, hybrids) supporting retrieval and accumulation.

7) Control Mode​

  • Human-in-the-Loop — Human must review/approve key decisions or outcomes before execution.
  • Human-on-the-Loop — Human supervises and can intervene/override, but the agent runs by default.
  • FullyAutomated — Operates without routine human oversight; bounded by policies/monitors/kill-switches.

8) Tool Use​

  • None — No external tools beyond built-in mechanisms.
  • FixedTools — Uses a predefined, declared set of tools/models/workflows; no runtime expansion.
  • DynamicToolUse — Can select, compose, or discover tools/models at runtime (e.g., function calling, plugins).
  • LegalPerson — Recognized as a legal person or represented by one (e.g., humans, incorporated entities).
  • NonLegalPerson — No inherent legal personhood (typical for software/robots/biological swarms).
  • ProxyForLegalPerson — Acts as an agent on behalf of a legal person (e.g., a bot executing an organization’s policies).

Notes on Interpretation​

  • Inheritance: Subclasses inherit facet meanings from ancestors; overrides narrow or fix values.
  • Tags: Simulated and Quantum are tags—they annotate the substrate context without replacing it.
  • Resolved Profiles: When presenting a node, show the resolved facet set (root ⊕ ancestor overrides ⊕ node overrides) and optionally badge entries overridden at that node.