AI Integration for OpenShift Assisted Installer

The integration surfaces within the installer's core workflow: the Infrastructure Environment definition, Host Discovery phase, and Cluster Configuration generation. AI agents ingest hardware inventory, network scans, and disk performance data from the installer's API to analyze against Red Hat's requirements and your organization's internal standards. This moves validation from pass/fail checks to predictive scoring—flagging potential issues like borderline RAM, disk latency that could impact etcd, or subnet conflicts before the installation job is even queued.

For implementation, an AI service acts as a pre-submission gate, calling the Assisted Installer's POST /v2/infra-envs and GET /v2/infra-envs/{infra_env_id}/hosts endpoints. It correlates host data with historical deployment logs to generate specific, actionable recommendations. For example: "Recommend increasing disk queue depth for host worker-03; 5 similar deployments experienced slow PVC provisioning." or "Detected mixed NIC models; suggest binding the SR-IOV network to the Intel XXV710 for consistent performance." These are appended to the cluster configuration as validated overrides, reducing post-install troubleshooting.

Rollout is phased, starting with a recommendation-only mode where AI insights are presented alongside installer validations for operator review. Governance is maintained through an audit trail linking each AI-suggested configuration change to the source data and model version used. This builds operator trust before progressing to automated approval workflows for low-risk changes, such as disk partitioning schemes or network MTU settings. The final integration connects to /integrations/kubernetes-and-container-management-platforms/ai-integration-for-openshift-gitops, enabling the AI-validated configuration to be directly committed as a GitOps source, closing the loop from intelligent analysis to declarative deployment.

The integration connects to the Assisted Installer's REST API and Event Stream at three key points: 1) During the cluster creation and host discovery phase, where AI analyzes hardware inventory and network validations; 2) At the pre-flight validation stage, where it interprets validation failures and suggests remediation; and 3) Post-installation, where it correlates installation logs with known failure patterns to generate root-cause analysis. The core data objects exchanged include Host inventory details, Cluster configuration manifests, validation Event payloads, and installation Log bundles. An AI agent acts as a middleware service, subscribing to installer webhooks, querying the API for real-time state, and posting back configuration recommendations or annotated troubleshooting guides.

In a production deployment, the AI service runs as a containerized workload on a management cluster, separate from the target bare-metal environment. It ingests structured validation results (e.g., "disk-speed insufficient") and unstructured log data, using a RAG pipeline grounded in Red Hat documentation, hardware compatibility lists, and historical deployment records. For governance, all AI-generated recommendations are logged with a confidence score and source citations, and can be routed through an optional approval workflow in a platform like ServiceNow or Jira before being applied via the Assisted Installer API. This ensures changes are auditable and can be reviewed by platform engineers.

Rollout typically follows a phased approach: starting with a read-only analysis mode that provides recommendations to engineers via a dashboard or Slack alert, then progressing to automated remediation for low-risk, high-confidence issues (e.g., suggesting a BIOS setting change), while blocking or escalating complex network or storage configuration changes. The integration's value is operational: reducing the time platform teams spend manually triaging pre-flight checks from hours to minutes, and decreasing installation rollbacks by providing actionable, context-aware troubleshooting before the installer proceeds to a likely failure state. For teams managing fleets of clusters, this AI layer becomes a force multiplier, codifying tribal knowledge and accelerating bare-metal provisioning cycles.

This architecture complements our broader Kubernetes platform integrations. For managing the deployed clusters, see our guides on AI Integration for OpenShift for operational workflows and AI Integration for OpenShift GitOps for post-installation configuration management.

AI agents interact with the Assisted Installer's REST API and must operate under strict identity and access management (IAM) principles. This involves using service accounts with scoped RBAC, limiting permissions to read-only analysis and configuration suggestion endpoints, while keeping actual cluster deployment and infrastructure modification actions gated behind human approval or existing GitOps workflows. All AI-driven API calls should be logged to the cluster's audit trail, correlating suggestions with the final, human-approved configuration that was applied.

A phased rollout is critical for managing risk and building trust. Start with a read-only analysis phase, where the AI agent reviews hardware inventory, pre-flight checks, and existing configuration manifests to generate reports and recommendations—with zero ability to apply changes. Next, move to a guided configuration phase, where the agent can generate draft install-config.yaml files or NMState configurations for review in a pull request. Finally, in a controlled automation phase, approved patterns (like standard SNO or compact three-node cluster layouts) can be fully automated, but only within pre-defined sandbox environments or for non-production clusters.

Security extends to the AI's operational data. Any data sent to an LLM for analysis (e.g., error logs, hardware details) must be scrubbed of sensitive identifiers. For on-premise or air-gapped deployments, the entire AI inference stack—including the vector store for historical failure analysis and the model itself—must be deployable within the disconnected environment. This ensures that hardware specifications, network topologies, and failure data never leave the secure perimeter, aligning with the security posture expected for managing Red Hat OpenShift infrastructure.