Create Custom AI System in Admin Console

Click “Create System” from the dashboard
Select “Custom” mode for full configuration
Select “Air-Gapped” for isolated environment with no external network access or “Normal” for standard deployment with external network access.

General Settings

System Name: Choose a unique name (e.g., pg-system-v3)
Namespace Configuration: Enable automated Kubernetes namespaces creation.
Pull Secret Configuration: “Enable Pull Secret” to use image pull secret for accessing private container registries. “Use Default Prediction Guard Credentials” to automatically create Kubernetes secret with the specified name.
OpenTelemetry (Otel) Config: Enter your OTEL Collector Endpoint to enable distributed tracing and observability for your AI system.
Advanced Scheduling: Configure Kubernetes scheduling constraints for all system components, including node selectors, tolerations and affinity rules.

Core Components

Core Service Component: Enable a centralized core service that orchestrates all system operations.
Redis Component: Leave blank for default Redis deployment or provide custom Redis connection details.
Authentication Service: Default of 1 replica
API Component: Default of 1 replica

Ingress

Public API Endpoint: Enable for external API access. Add TLS hosts and TLS secret name
Nginx Component: Enable for reverse proxy and load balancing

Data Processing

Data Processing Use Default for all resources for standard deployment.

Pre-Processing: Enable for pre-processing tools like PII anonymization and Prompt Injection Detection.
Post-Processing: Enable for post-processing tools like factuality check and toxicity filtering.
Document Processing: Enable for document processing for extraction and analysis of text from various file formats (PDFs, images, Word documents)
Audio Processing: Enable for audio processing for transcription and analysis of audio files. Also supports audio diarization.

Operations

Cert Manager: Enter your Cert Manager configurations to automate the provisioning and renewal of SSL/TLS certificates for your system. It integrates with certificate authorities like Let’s Encrypt to ensure secure HTTPS connections without manual certificate management.

If you enabled a Public API Endpoint with TLS hosts in the Ingress section, make sure your DNS records for those hostnames point to the NGINX ingress IP address on your cluster. Without this, external traffic will not route correctly to your AI system. To do this, you can run kubectl get service ingress-nginx-controller -n predictionguard to get the external IP address of the NGINX ingress controller and then update your DNS records accordingly.

Review

Review your configurations and click “Create System” to register your system and generate the deployment command. You can then use this command to deploy your system in your environment.