Automated Setup

setup-maas.sh is a single script that handles the entire MaaS lifecycle, from operator installation through model deployment and verification. Each phase is idempotent; re-running skips what’s already done.

This guide is not a replacement for the official Red Hat OpenShift AI Models as a Service documentation. It is a companion resource with opinionated Kustomize manifests and automation scripts to accelerate deployment.

Clone and Run

git clone https://github.com/rh-aiservices-bu/rhoai-maas-guide.git
cd rhoai-maas-guide
./scripts/setup-maas.sh

This runs Phases 0-6 automatically. The script detects what’s already installed and skips completed phases.

Arguments

Option Description Default

--model <name>

Model to deploy: simulator (CPU), granite-tiny-gpu (small GPU), gpt-oss-20b (large GPU), auto (auto-detect by GPU VRAM)

auto

--from-phase <N>

Start from phase N (0-7), skipping earlier phases

0

--skip-models

Skip Phase 5 (model deployment)

--skip-verify

Skip Phase 6 (verification)

--with-observability

Also run Phase 7 (COO + Gateway telemetry)

--dry-run

Preview what would be applied without making changes

Phases

Phase What it does Time

0

Preflight - detect cluster state, decide which phases to run

instant

1

Operators - oc apply -k manifests/01-prerequisites/operators/, wait for CSVs

2-5 min

2

Platform - Kuadrant, UWM, GatewayClass, envsubst Gateway

2-5 min

3

MaaS Platform - PostgreSQL secrets/deployment

2-3 min

4

RHOAI - DSC with modelsAsService: Managed, Dashboard flags, wait for maas-api

3-5 min

5

Model - auto-detect GPU, apply Kustomize, wait for Ready

0.5-15 min

6

Verify - 6-phase E2E (API, auth, rate limits, cleanup)

3-5 min

7

Observability - COO subscription + Gateway TelemetryPolicy

2-3 min

Common Usage Patterns

Full installation (recommended)

For a fresh cluster with no MaaS components:

./scripts/setup-maas.sh

With observability

./scripts/setup-maas.sh --with-observability

Resume from a specific phase

If a previous run failed at Phase 4:

./scripts/setup-maas.sh --from-phase 4

Models only

If the platform is already installed and you just want to deploy a model:

./scripts/setup-maas.sh --from-phase 5

Or use the standalone model script for more control:

./scripts/deploy-model.sh --model simulator

Verify only

./scripts/setup-maas.sh --from-phase 6

Or directly:

./manifests/06-verification/verify.sh

Convenience Scripts

These wrap individual phases for standalone use:

Script What it does

scripts/deploy-model.sh

Phase 5 only - deploy a model with GPU auto-detection

scripts/test-inference.sh

Send inference requests to a deployed model endpoint

scripts/verify-maas.sh

Phase 6 only - wrapper for manifests/manifests/06-verification/verify.sh

State Detection

Phase 0 inspects the cluster and reports what’s already installed. The script uses this to skip completed phases:

  • Operators installed? Skip Phase 1.

  • Kuadrant + UWM + Gateway ready? Skip Phase 2.

  • PostgreSQL running? Skip Phase 3.

  • DSC with modelsAsService: Managed + maas-api running? Skip Phase 4.

  • Models deployed in llm namespace? Skip Phase 5.

Override with --from-phase N to force re-running from a specific phase.

AI-Assisted Installation

If you use an AI coding tool (Claude Code, Open Code, Cursor, Roo Code), the /install-maas skill wraps the same script with an interactive, AI-assisted workflow. Clone the repo, open it in your tool, and run:

/install-maas

Or with arguments:

/install-maas --model simulator --with-observability

See AI-Assisted Installation for details.

Final Report

The script ends with a summary: RHOAI version, MaaS API URL, Gateway status, health check, deployed models, and suggested next steps.