Skip to main content

​
Agent Abstraction Triad

Each part answers a different question. The behavior describes what the agent can do. The environment describes how that behavior runs in a specific project context. The identity describes who or what is being addressed when the environment is exposed as a direct agent.
  • Agent behavior is the executable logic and capability declaration.
  • Agent environment is the configured runtime context that resolves and runs a behavior.
  • Agent identity is the stable address and permission subject for direct access.
Aion keeps these concerns separate so the same implementation can be reused, configured, addressed, and authorized in different ways.

​
Agent Behavior

Agent behavior is the implementation. It describes the executable logic, lifecycle kind, configuration schema, capability declarations, and optional agent card for a versioned behavior. A behavior can be backed by an Aion deployment, a remote A2A deployment, a distribution behavior, or a system behavior. A behavior does not, by itself, decide where it runs, which release channel is active, which configuration values are supplied, or which identity is used to address it. Those choices belong to the environment and identity. This separation lets one behavior appear in several roles:
  • A terminal agent in a distribution sequence
  • Middleware inside a sequence
  • A direct daemon agent exposed for internal agent-to-agent calls
See Behaviors for the behavior-specific model.

​
Agent Environment

Agent environment is the configured runtime context for a behavior. An environment belongs to a project and carries the settings needed to resolve and run a behavior. It stores the behavior lineage to use, the runtime selector for stable, beta, or pinned behavior resolution, environment-specific configuration variables, and capability configuration. The environment is also the place where Aion binds a daemon identity. When an environment has a daemon identity, Aion can resolve direct requests from that identity back to the environment, then resolve the environment to the behavior that should run. This means the environment is the bridge between reusable behavior and a specific, addressable agent instance. See Agent Environments for the environment details.

​
Agent Identity

Agent identity is the stable address and permission subject for direct access. In this abstraction, the identity is the daemon identity bound to an environment. It represents an internal component principal, anchors roles and permissions, and lets other agents address the component directly. Aion has other identity types. Distribution identities describe public ingress, personal identities represent users, and system identities represent platform-managed actors. The daemon identity is what makes a configured environment directly callable as a singular agent. See Identities and Daemons for the broader identity and daemon models. At runtime, a daemon request resolves from identity to environment to behavior. Aion then applies capability configuration, authorization, and daemon context before invoking the behavior.

​
Scope

​
Nodes

Nodes are the internal representation of agents within a project. Every node in the project editor is effectively an agent participant. It has an environment, resolves a behavior, and may expose a daemon identity for direct access from other agents. This is why daemon identities are bound to environments: the identity addresses the node’s direct agent surface, and the environment resolves the behavior that runs. In that model, nodes are like cells in a larger organism. Each node is an agent on its own, and composed nodes can form a larger agent system. See Nodes for the project-node model.

​
Distributions

Distributions are the outward-facing representation of an agent system. A project can compose many internal nodes, but it still needs a surface that represents the project as a single unit to the outside world. A distribution provides that surface. It connects external requests to the project’s internal agent graph and returns responses through the same public route. In that sense, the distribution is how the composed agent system is reflected outside the project. The internal nodes act as agent participants, while the distribution presents the larger composed agent through an external connection. See Distributions for the public routing model.