Skip to content

API Reference

While Agentic-Notify exposes many sub-modules, these are the core classes you will interact with while building your orchestration pipelines.

NotificationOrchestrator

The central coordinator class (agentic_notify.orchestrator). It manages the registration of Adapters, Handlers, and Workflows, and exposes the primary event processing loop.

  • register_adapter(adapter: BaseAdapter): Mounts an external tool for execution.
  • register_handler(handler: BaseHandler): Mounts an internal logic handler (like Human-in-the-loop).
  • register_workflow(workflow: WorkflowDefinition): Stores a static execution graph.
  • async process_event(raw_payload: Dict[str, Any]) -> Dict[str, Any]: The main ingestion pipe. Accepts an unstructured dictionary, norms it, and executes the lifecycle.

AgenticPlanner

The LLM integration bridge (agentic_notify.planning.planner). It generates multi-step workflows dynamically.

  • async plan_workflow(intent: str) -> WorkflowDefinition: Takes a natural language string, prompts the LLM alongside the loaded ToolRegistry, and strictly attempts to validate the returned JSON string into a Pydantic WorkflowDefinition model.

NotificationEvent

The canonical data model for all inbound signals (agentic_notify.schemas.notification).

class NotificationEvent(BaseModel):
    event_id: str
    source_platform: str
    source_app: str
    title: Optional[str] = None
    body: Optional[str] = None
    received_at: datetime
    metadata: Dict[str, Any] = {}

WorkflowDefinition

The Pydantic schema mapping the Execution Graph (agentic_notify.schemas.workflow).

class WorkflowDefinition(BaseModel):
    workflow_id: str
    workflow_name: str
    steps: List[WorkflowStep]
  • WorkflowStep: The core state-block defining how standard data inputs map to an Adapter's expected arguments. Features id, kind, name, and input_from mappings.