main.py

"""
Hatchet alternative - submit an ML training pipeline with durable delivery.

Hatchet needs workers, task queues, and workflow engine infrastructure.
AXME needs one intent.

Usage:
    pip install axme
    export AXME_API_KEY="your-key"
    python main.py
"""

import os
from axme import AxmeClient, AxmeClientConfig


def main():
    client = AxmeClient(
        AxmeClientConfig(api_key=os.environ["AXME_API_KEY"])
    )

    # Submit an ML training task - replaces Hatchet worker + task queue + workflow
    intent_id = client.send_intent(
        {
            "intent_type": "intent.tasks.run.v1",
            "to_agent": "agent://myorg/production/task-runner",
            "payload": {
                "task_id": "TASK-2026-5567",
                "task_type": "ml_training_pipeline",
                "model": "fraud-detector-v3",
                "dataset": "transactions-2026-q1",
                "epochs": 50,
            },
        }
    )
    print(f"Intent submitted: {intent_id}")

    # Observe lifecycle events in real time (SSE stream, no polling)
    print("Watching lifecycle...")
    for event in client.observe(intent_id):
        status = event.get("status", "")
        print(f"  [{status}] {event.get('event_type', '')}")
        if status in ("COMPLETED", "FAILED", "TIMED_OUT", "CANCELLED"):
            break

    # Fetch final state
    intent = client.get_intent(intent_id)
    print(f"\nFinal status: {intent['intent']['lifecycle_status']}")


if __name__ == "__main__":
    main()