Deployments

Put your AI model to work. A Deployment connects your trained model to live Zabbix data to detect and report anomalies as they happen.

When you deploy a model, Prisma analyzes your incoming data and sends the resulting anomaly score back to your Zabbix.

Create a deployment

To move a AI model to your production environment:

  • Navigate to AI > Deployments.
  • Click the + (plus) button in the bottom-right corner.

Step 1: General

  • Name: Choose a name for your AI model deployment.
  • Model configuration: Select the model configuration you want to use.
  • Trained model: Choose the specific trained AI model you wish to use.

Step 2: Zabbix integration

To see the AI’s findings within Zabbix, you must define where the results are sent:

  • Output Zabbix host: Select the Zabbix host where Prisma should create the anomaly result items.
  • Item key prefix: Enter a short text string to prepend to the Zabbix item keys. This ensures that if you use the same output host for multiple deployments, the data remains unique and organized.

Click Deploy. Prisma will now automatically generate the necessary items and triggers on your selected Zabbix host.

How Prisma reports anomalies

Once active, the deployment monitors your data and calculates two distinct levels of concern, which appear as items in Zabbix:

  • Potential anomaly Score: Indicates unusual behavior with lower confidence. These are “heads-up” alerts for early warnings.
  • Significant anomaly Score: Indicates a high-confidence anomaly. These represent deviations that the AI is certain are abnormal.

Prisma also creates Zabbix triggers automatically. These will fire a “Problem” on your Zabbix server whenever an anomaly score exceeds the safety threshold.

Managing your deployment

Click on a deployment in the list to open its details page and manage its live operation.

Status & General

On the Deployment details page you can:

  • View the current status of the deployment, and temporarily pause or resume the deployment at any time.
  • Change the name or description of the AI deployment.
  • See the Zabbix output host, where the deployment sends its anomaly scores to.

Adjust sensitivity

Every environment has a different “noise” level. You can fine-tune how easily the AI triggers an alert by changing the Sensitivity setting.

  • High confidence only: The most strict setting; only reports the most obvious anomalies.
  • Conservative: Fewer alerts, focusing on clear deviations.
  • Recommended: The recommended balanced default setting for most environments.
  • Aggressive: Reports even slight deviations.
  • Very aggressive: The most sensitive setting; alerts on almost any change from the norm.

After changing the sensitivity, please wait a few moments for the new logic to be synchronized and applied to your Zabbix triggers.

Monitor live anomalies

For every Zabbix item included in your deployment, Prisma provides a dedicated data graph showing recent history.

  • Real-time data: The graph displays the actual value history of the selected Zabbix item.
  • Anomaly markers: If the AI detects a deviation from the norm, it will highlight that specific moment directly on the graph:
    • Yellow markers: Indicate a potential anomaly. These are early warnings where the AI sees unusual behavior but with lower confidence.
    • Red markers: Indicate a significant anomaly. These are high-confidence deviations that represent a clear break from normal patterns.