Skip to main content

​
Introduction

This procedure explains how to deploy and operate the Autonomous Retail Billing Agent on a node running EVE-OS (Edge Virtualization Engine). The agent processes real-time vision data locally to identify retail items and calculate bills, ensuring privacy and low latency without requiring a constant connection to Edge Infrastructure Services. You can control the full retail self-checkout billing lifecycle through the agent’s chat interface.

​
Prerequisites

  • You have a role that can submit deployment requests: Organization Admin, MLOps Admin, or MLOps Engineer in your organization, or the Super Admin system role. See Organizations, Users, and Roles.
  • A ZEDEDA Edge Kubernetes Service cluster is imported and running. For setup instructions, see the ZEDEDA Edge Kubernetes Service quick start.
  • The target node is a Jetson AGX Thor device or Developer Kit (recommended for default deployment).
  • The target node has one NVIDIA GPU available to the agent, at least 16 GB of memory (up to 32 GB used), and storage for about 60 GB of persistent volumes (model cache and application data).
  • Physical hardware is available: a tray with retail items, a camera connected to the node, and a connected speaker.
  • A retail item database with current catalog items and prices is available and accessible to the agent.
  • You have a Hugging Face access token. The agent requires it for every deployment, not only for gated models, and does not start without it. Enter it as the Hugging Face token value in the deployment wizard’s configuration values.

​
Deploy the Retail Billing Agent

  1. In the sidebar, select Marketplace > AI Apps.
  2. Locate the Autonomous Retail Billing Agent in the gallery. Use the search field if needed.
  3. Select the agent card to open the agent detail page.
  4. Select the Deploy icon.
  5. Follow the deployment wizard to select the target ZEDEDA Edge Kubernetes Service cluster and complete the deployment configuration.
Deploying creates a deployment request with status Pending. Organization Admin or MLOps Admin in that organization must approve it; users cannot approve their own requests. See Approve and Validate Models. After the request is approved and deployment completes, the agent starts on the node and its chat interface becomes available. On first start, the agent downloads its model from Hugging Face, which requires internet access and can take 10 minutes or more.

​
Run the Autonomous Retail Billing Agent

  1. Open the agent’s web interface by navigating to the Agent Chat URL for your node.
  2. In the Agent Chat, select Live Preview in the navigation bar.
  3. Place the retail items on the tray underneath the camera.
  4. Enter the command scan the tray and count the items to identify the products.
  5. Enter the command create a bill and generate an invoice to calculate the total cost.
  6. Listen for the total cost through the connected speaker.
  7. Enter the command send the invoice to [email address] to deliver the invoice to the customer.
  8. Approve the action prompt to authorize the agent to send the email.
  9. Select Logs to review the invoice log entries.

​
Verify the Configuration

  • Verify that the agent correctly identifies the items shown in the chat and updates the invoice total.
  • Check the target email inbox to confirm that the generated invoice PDF was delivered.

​
In This Section