Intel® AI for Enterprise Inference Documentation
The technical documentation for the inference layer. Start with Deploy to get a model serving, use Customize to add models and tune the engines, then look things up in Reference.
New here? The three-step quick start is in the project README. This layer is a component of Intel® AI for Enterprise Solutions — install that platform first.
Deploy
Get a model running and call it.
| Guide | What it covers |
|---|---|
| Deploy a Model | Every model-manager command and flag — deploy, update, undeploy, gated models, ad-hoc Hugging Face deploys |
| Accessing Models | Endpoints and credentials for both auth modes, and how to list what's serving |
Customize
Add models, choose engines and versions, and change any setting.
| Guide | What it covers |
|---|---|
| Model Catalog | The models.yaml catalog — every field, and auto-deploy at install time |
| Runtimes | vLLM and OpenVINO — versions, images, and per-workload defaults |
| Configuration | Every option in config.yaml, plus the MM_* environment variables |
Reference
How the layer is built and how a request travels through it.
| Guide | What it covers |
|---|---|
| Architecture | Components, Ansible roles, and how this repo plugs into the platform |
| Inference Request Flow | End-to-end request routing for both auth modes, with a verified network trace |
| CPU Pinning & NUMA | The three CPU policies, the pre-flight capacity check, and NUMA placement |
| Database Architecture | Shared PostgreSQL schema and provisioning for LiteLLM and Langfuse |
Support
| Guide | What it covers |
|---|---|
| Troubleshooting | Common symptoms and their fixes |