Datasets / Agentic AI Lexicon
Agentic AI Lexicon
A lexicon of agentic AI concepts — agents, orchestration, identity, policy and execution layers.
Machine-readable downloads: JSON · CSV
Starter edition (20 terms). The full lexicon targets ~300 terms; expand incrementally.
- Plan-Act-Observe Loop
- The core agent cycle: form a plan, execute an action, observe the result, and iterate.
- Multi-Agent System
- A system where several agents coordinate, negotiate or compete to achieve goals.
- Agent Identity
- A cryptographically verifiable identity assigned to an autonomous agent.
- Action Mediation
- Intercepting an agent's intended actions to check them against policy before execution.
- Execution Layer
- The controlled runtime in which an agent's actions are mediated and logged.
- Policy Layer
- The layer that defines and evaluates the deterministic rules governing agent behavior.
- Identity Layer
- The layer that establishes and verifies who an agent is acting as.
- Audit Layer
- The layer that records an immutable trail of agent decisions and actions.
- Capability Token
- A scoped credential granting an agent a specific, time-bound permission.
- Delegation
- Granting one agent the authority to act on behalf of a user or another agent.
- Escalation Policy
- Rules defining when an action must be routed to a human or higher authority.
- Blast Radius
- The maximum potential impact of an agent action if it goes wrong.
- Containment
- Limiting an agent's reach so failures cannot propagate across systems.
- Deterministic Checkpoint
- A fixed control point where the same conditions always yield the same governance decision.
- Agent Registry
- A catalog of known agents, their identities, roles and permitted capabilities.
- Context Window
- The span of information an LLM can consider at once when reasoning.
- Grounding
- Anchoring an agent's outputs to verified data or sources to reduce hallucination.
- Hallucination
- A confident but unsupported or fabricated output from a model.
- Guard Agent
- A supervisory agent that monitors and constrains other agents.
- Orchestrator
- The component that assigns tasks, sequences agents and enforces shared policy.
Compiled by Gianluca Busato, Enkronos ecosystem. Free to cite with attribution.