Product Documentation for Red Hat OpenShift AI Self-Managed 3.4

Overview of the new features included in the 3.4 Early Access (EA1) release

Install via Operator or CLI, enable required components, verify the deployment, and cleanly uninstall when needed

Install via Operator or CLI, enable required components, verify the deployment, and cleanly uninstall when needed

Get set up to create projects, launch workbenches, and deploy your first model on OpenShift AI

Review compatibility matrices, accelerator support, deployment targets, and update policy prior to installation

Explore the curated set of third‑party models validated for Red Hat AI products, ready for fast, reliable deployment

Use CRDs or dashboard to publish images and provision resourced workbenches

Administer access, apps, resources, and accelerators; maintain logging, audit, and backups

Use Feature Store to define, store, and serve reusable machine learning features to models

Help administrators decide what usage data is collected, see what’s included, and enable or disable telemetry

Enable supported hardware configurations for your workloads

Configure your model-serving platform in Red Hat OpenShift AI Self-Managed

Operate Llama Stack: activate the operator and expose OpenAI‑compatible RAG APIs

As an administrator, configure user access, customize the dashboard, and manage specialized resources for data science and AI engineering projects

Use the OpenShift AI dashboard to create registries, set access with RBAC groups, and manage model and version lifecycle so teams can register, share, and promote models to serving with traceability

Plan which APIs to build on and how to upgrade with minimal risk by mapping each OpenShift AI endpoint to a support tier that defines stability and deprecation timelines

Store, version, and promote models with metadata for cross‑project sharing and traceability

Use the model catalog to discover, evaluate, register, and deploy models for rapid customization and testing

Enable LlamaStack, GPUs, and vLLM, ingest data in a vector store and expose secure endpoints

Experiment with models in the gen AI playground in Red Hat OpenShift AI Self-Managed

Distribute data and ML jobs for faster results, larger datasets, and GPU‑aware auto‑scaling and monitoring

Create a connection, configure an S3 client, and list, read, write, and copy objects from notebooks

Organize projects, collaborate in workbenches, build notebooks, train/deploy models, and automate pipelines

Launch a workbench, pick an IDE, and develop with prebuilt images or custom environments

Define KFP‑based pipelines, version and schedule runs, and track artifacts in S3‑compatible storage

Enable applications, connect with keys, remove unused tiles, and access Jupyter from the dashboard

Configure LMEvalJobs, select tasks, run evaluations, and retrieve metrics to compare model performance

Orchestrate detectors to filter LLM inputs/outputs, auto‑configure security, and expose guarded endpoints

Monitor model bias and data drift by configuring metrics, thresholds, and visualizations in OpenShift AI

Deploy models with KServe—choose RawDeployment or Knative, set resources and runtimes, and expose authenticated endpoints

Govern LLM access with Models-as-a-Service in Red Hat OpenShift AI Self-Managed