AgentConfig

Kind: interface

Signature

export interface AgentConfig

Remarks

steps are validated structurally here because authored step functions live in TypeScript, not in JSON-serializable manifest data.

Properties

• identity (AgentIdentity): Identity and scope information for the agent.
• tools (string[]): Tool binding identifiers the agent may invoke.
• data (AgentDataConfig): Data access declarations used by governance and the authoring UI.
• triggers (AgentTriggerDefinition[]): Trigger definitions that can start runs for the agent.
• manualInput? (ManualInputDefinition): Optional manual-run contract edited in code and the GUI.
• chatOutput? (ChatOutputDefinition): Optional chat-facing output selector for chat-triggered runs.
• dlp? (DlpScanPolicyConfig): Optional agent-level DLP defaults and action-family overrides.
• workflow (string[]): Declared workflow labels or edges for the agent.
• recovery? (WorkflowRecoveryDeclaration): Workflow-authored restrictions on connector-attested recovery behavior.
• trust (AgentTrustConfig): Initial trust posture assigned at deployment time. Authoring helpers default omitted trust declarations to { initialTier: "medium" }.
• memory (AgentMemoryConfig): Optional scoped-memory defaults. Memory is enabled when omitted.
• localRules (AgentLocalRule[]): Optional agent-scoped governance rules authored directly in source control.
• steps (unknown[]): Step definitions attached by the authoring DSL.
• kind? (AgentKind): Execution mode. Defaults to workflow; autonomous agents run one platform-generated step that drives kernel.agentLoop.
• loop? (AgentLoopConfig): Autonomous loop declaration. Required when kind is autonomous; also usable by workflow agents that call kernel.agentLoop inside a step.