Automate Tier-1 Support with the Support Resolver Agent

1. 1
   Connect Your CRM

   Grant the Support Resolver Agent read/write access to your CRM (e.g., Salesforce, HubSpot). This is essential for fetching customer history and logging resolution actions, enabling personalized and context-aware support.
2. 2
   Link Your Payment Gateway

   Connect your payment provider (e.g., Stripe) to allow the agent to process refunds and manage subscriptions. This is the core of the agent's ability to take direct transactional action instead of just answering questions.
3. 3
   Set Action Policies and Safety Caps

   Define the agent's operational boundaries. Set a maximum monetary value for refunds (e.g., $75) and specify which subscription plans it can modify. This is a critical safety feature to prevent unintended high-value actions.
4. 4
   Configure Smart Escalation

   Determine where unresolved cases are routed. Configure a webhook URL for your human support desk (e.g., Zendesk, Intercom) so the Support Resolver Agent can escalate issues with a full conversation summary and diagnostic data.
5. 5
   Test with a Refund Request

   Send a sample user request to your deployed agent's endpoint to verify its behavior. A common test case is a refund request that falls within the configured monetary cap. Check your payment gateway to confirm the refund was processed.

import requests
import json

# --- Configuration ---
API_BASE_URL = "https://api.agents-as-a-service.com/v1"
API_KEY = "YOUR_API_KEY"  # Replace with your actual API key
AGENT_ID = "YOUR_AGENT_ID" # Replace with your Support Resolver Agent ID

HEADERS = {
    "Authorization": f"Bearer {API_KEY}",
    "Content-Type": "application/json"
}

def configure_agent():
    """Connects CRM, payment gateway, and sets policies for the agent."""
    print("1. Configuring Payment Gateway (Stripe)...")
    stripe_config = {
        "provider": "stripe",
        "credentials": {"apiKey": "sk_test_YOUR_STRIPE_API_KEY"}
    }
    requests.post(f"{API_BASE_URL}/agents/{AGENT_ID}/connections/payment", headers=HEADERS, data=json.dumps(stripe_config)).raise_for_status()
    print("   ...Done.")

    print("2. Setting Action Policies...")
    policy_config = {
        "refunds": {"maxAmount": 75, "currency": "USD"},
        "subscriptions": {"allowModification": ["plan_pro"]}
    }
    requests.put(f"{API_BASE_URL}/agents/{AGENT_ID}/policy", headers=HEADERS, data=json.dumps(policy_config)).raise_for_status()
    print("   ...Done.")

    print("3. Setting Escalation Webhook...")
    escalation_config = {
        "type": "webhook",
        "url": "https://hooks.your-support-desk.com/services/XYZ"
    }
    requests.put(f"{API_BASE_URL}/agents/{AGENT_ID}/escalation", headers=HEADERS, data=json.dumps(escalation_config)).raise_for_status()
    print("   ...Done.")

def test_agent_refund():
    """Sends a test conversation to the agent to trigger a refund."""
    print("\n--- Testing Agent with Refund Request ---")
    test_payload = {
        "userId": "customer_abc_123",
        "text": "Hi, my last order #54321 was incorrect. I'd like a refund please."
    }
    try:
        response = requests.post(f"{API_BASE_URL}/agents/{AGENT_ID}/converse", headers=HEADERS, data=json.dumps(test_payload))
        response.raise_for_status()
        print("Agent Response:", response.json().get('response', 'No response text.'))
        print("Action Taken:", response.json().get('action', 'No action taken.'))
    except requests.exceptions.HTTPError as e:
        print(f"Error: {e.response.status_code} - {e.response.text}")

if __name__ == "__main__":
    # Note: In a real scenario, you'd run configure_agent() once during setup.
    # configure_agent()
    test_agent_refund()