The fast path to deploying enterprise AI on Intel® Xeon® silicon. Go from bare metal to a working AI stack in minutes.
Deploy and connect LLM inference, RAG, and agentic workflows in a customizable Kubernetes-based platform — with a GenAI gateway, security, intelligent routing, observability, model serving, and infrastructure automation built in.
Enterprise AI needs more than a model endpoint. Teams need to connect LLM inference, RAG, and agentic workflows, along with security, routing, and observability.
Intel® AI for Enterprise Solutions wires these pieces together in a pre-integrated, customizable Kubernetes stack optimized for Intel® Xeon® processors.
A single installer script, es_auto_installer.sh, takes bare-metal nodes through Kubernetes provisioning, platform services (networking, security, observability), model serving, and AI gateway setup — delivering a production-ready AI platform with OpenAI-compatible endpoints in one run.
The end result: teams get a secured, load-balanced stack where they can immediately serve LLM models, run RAG pipelines, and monitor everything through a unified gateway. Start with defaults, then configure or extend as needed.
Want the full picture? See Architecture and Meet AI for Enterprise Solutions.
The stack installs in four ordered layers, in the order they are deployed:
| # | Layer | Components | Source | Documentation |
|---|---|---|---|---|
| 1 | Infrastructure | Kubernetes (via Kubespray) · Storage (local-path · NFS · Ceph · NetApp ONTAP) · Intel® Xeon® | enterprise-ai-solutions | Deployment Guide |
| 2 | Platform | Istio ambient mesh, Envoy Gateway, PostgreSQL, Keycloak, MinIO, observability | enterprise-ai-solutions | Deployment Guide |
| 3 | Inference | Envoy AI Gateway, KServe, vLLM / OpenVINO™ Model Server — exposes the model endpoint on top of which services like RAG and agents can be built | enterprise-inference | Deploy a Model |
| 4 | Intel® AI for Enterprise RAG (opt-in) | Vector database, document ingestion (EDP), RAG pipeline orchestration (GMC), MCP gateway, chat history, web UI — requires init erag and install erag |
enterprise-rag | Getting Started with RAG |
Request flow: a request enters through the Envoy AI Gateway, is authenticated against Keycloak (or a LiteLLM virtual key), and is routed to the matching model-serving backend. Every layer's health and latency is visible in the built-in Grafana / Prometheus / Loki / Tempo stack.
See the Architecture deep-dive for the full component list, execution flow, and cross-repo layering.
One node, defaults, and roughly 20 minutes from bare metal to a working stack. Each target pulls in the layers below it automatically, so the target name is the only thing that changes between use cases.
Note
Prerequisites: Ubuntu 22.04/24.04 (or RHEL/Rocky x86_64), passwordless sudo, and internet access. Full list → Prerequisites.
| You want | Target | Then follow |
|---|---|---|
| An OpenAI-compatible LLM endpoint to build on | inference |
Deploy a Model |
| A complete RAG application: document ingestion, vector search, chat UI | erag |
Getting Started with RAG |
| To connect your own app, agent, or framework to the endpoint | inference |
Integration Guide |
git clone https://github.com/intel/enterprise-ai-solutions.git
cd enterprise-ai-solutions
./es_auto_installer.sh configure # one-time machine prep (Python 3.11+, yq, kubectl, helm)
./es_auto_installer.sh init inference # swap in your target from Step 1
./es_auto_installer.sh install inference # deploys the target plus everything it depends on (~15-20 min)Point kubectl at the new cluster:
export KUBECONFIG=$(pwd)/env/local/kubeconfig.yaml
kubectl get nodes # should show ReadyAll settings live in env/local/global_config.yaml, created by init. Defaults are a single local node with Keycloak OIDC auth. See Configuration Reference for every option, and Deployment Guide for multi-node, bastion, bring-your-own-cluster, and the alternative litellm virtual-key auth mode.
Tip
--env defaults to local. Install and teardown are environment-scoped: what you installed with --env prod must be torn down with --env prod. Remove everything (config preserved) with ./es_auto_installer.sh teardown infrastructure.
model-manager downloads the weights, picks serving parameters, and creates the pod - no YAML needed:
./model-manager deploy qwen3-0-6b --waitThe helper script discovers the gateway address and Keycloak credentials from the cluster and exports a JWT:
source ./ext/enterprise.ai-inference/model_manager/scripts/get-keycloak-token.sh
# Exports: TOKEN, GATEWAY_IP, GATEWAY_DOMAIN
curl -sk --noproxy '*' \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
--resolve "$GATEWAY_DOMAIN:443:$GATEWAY_IP" \
-d '{"model":"qwen3-0-6b","messages":[{"role":"user","content":"Hello!"}],"max_tokens":64}' \
"https://$GATEWAY_DOMAIN/v1/chat/completions"Everything speaks the OpenAI-compatible API, so any client that works with OpenAI works here. Gated models, ad-hoc Hugging Face deployments, the Python SDK, and the litellm access flow are covered in Deploy a Model. Building the RAG application instead? Continue with Getting Started with RAG.
The Quick Start gets you running with single-node defaults. From here you can tailor almost everything — TLS and authentication, multi-node or bring-your-own cluster, storage backends, RAG pipelines, model catalogs, autoscaling, proxies, and which components to enable or swap.
| Goal | Guide |
|---|---|
| All configuration options | Configuration Reference |
| Multi-node cluster or BYO Kubernetes | Topologies |
| Deploy and manage models | Deploy Models |
| Deploy the RAG layer | Getting Started with RAG |
| Connect your app or framework | Integration Guide |
| CLI commands and flags | CLI Reference |
| Architecture deep-dive | Architecture |
| Common questions | FAQ |
| Terminology | Glossary |
| Network topology and ingress | Network Architecture |
| Trident + ONTAP | NetApp ONTAP and Trident |
| Workload placement (multi-node) | Node Topology |
| NUMA-aware CPU pinning | NRI CPU Balloons |
Licensed under the Apache License, Version 2.0.
- Documentation Index
- GitHub Repository
- Intel® Enterprise for AI Inference (inference layer)
- Intel® AI for Enterprise RAG (RAG layer)
- Meet AI for Enterprise Solutions
- Architecture
Intel®, Intel® Xeon®, and Intel® Arc™ are registered trademarks of Intel Corporation or its subsidiaries. Licensed under the Apache License, Version 2.0.