neuron-workflow

Neuron AI Workflow

This skill helps you build custom event-driven workflows in Neuron AI. Workflows are the foundation of the entire framework - Agent and RAG are built on top of Workflow.

Setters configure the reusable definition, including while a segment is running. An execution retains its resolved graph, resources, completion policy and event dispatcher. New listener registrations apply to subsequent segments. A call made while a segment is in flight meets the persisted run as another process would: the running generation refuses a new ignition.

Core Concepts

Event-Driven Architecture

Workflows operate through events flowing between nodes:

StartEvent → Node1 → Event2 → Node2 → Event3 → Node3 → StopEvent

Each node:

1. Receives a typed Event
2. Processes it
3. Returns a new Event (or StopEvent to complete)

The Node Pattern

Nodes extend the Node base class:

use NeuronAI\Workflow\Events\Event;
use NeuronAI\Workflow\Events\StartEvent;
use NeuronAI\Workflow\Events\StopEvent;
use NeuronAI\Workflow\Node;
use NeuronAI\Workflow\WorkflowState;

class ValidationNode extends Node
{
    // The __invoke signature determines which event this node handles
    public function __invoke(StartEvent $event, WorkflowState $state): ProcessEvent
    {
        $input = $state->get('input');
        $validated = $this->validate($input);
        $state->set('validated', $validated);
        return new ProcessEvent($validated);
    }

    protected function validate(mixed $input): array
    {
        // Validation logic
        return ['valid' => true, 'data' => $input];
    }
}

Key Pattern: The workflow automatically maps events to nodes based on the first parameter type of __invoke().

Defining Custom Events

use NeuronAI\Workflow\Events\Event;

class UserValidatedEvent implements Event
{
    public function __construct(
        public readonly string $userId,
        public readonly array $userData
    ) {}
}

class ProcessCompleteEvent implements Event
{
    public function __construct(
        public readonly string $result
    ) {}
}