---
title: "AI Model Center"
description: "Deploy your first on-premises model on ARX — pick a recommended model in Model Hub, check it in Model Instances, connect an app to it, and watch GPU usage"
canonical_url: "https://docs.ardge.com/en/guide/use-ai/ai-model-center"
---
# AI Model Center
> Deploy your first on-premises model on ARX — pick a recommended model in Model Hub, check it in Model Instances, connect an app to it, and watch GPU usage
## Overview
AI Model Center is where ARX manages AI models. Before an app can use local AI, you usually have to download and start the model it needs here.
**Find a model → download it → start it as a usable AI service → watch the GPU and model status**
The first time, pick from **Ardge Select** — models ARDGE has verified. They ship with recommended deployment parameters, so you don't need to understand GPU settings yet.
## Step 1: Choose where the model comes from
### Option A: Ardge Select (recommended to start)
Go to **Model Hub > Ardge Select**. Every model there is verified by ARDGE and ships with deployment parameters already set, so it runs without any tuning. Pick one and click **Download and Deploy** — the system downloads the model and starts it when it finishes.
**Download** without **Deploy** is also available. The model lands in **Downloaded Models**, where you can deploy it later.
### Option B: Hugging Face (advanced)
If you already know which Hugging Face model you want, search for it and deploy it directly. On its card, click **Download and Deploy** to open a three-step wizard:
1. **Download Settings** — confirm the model format and file
2. **Deploy** — set the deployment parameters. If you're not sure, leave **Deployment Configurations** on **Default**
3. **Resource Check** — the system verifies VRAM, storage and download permission
The wizard's own **Download and Deploy** button starts the job. The download runs in the background and the model deploys as soon as it finishes.
**A gated model needs a Hugging Face token:**
Some models require you to accept the provider's terms before downloading. The **Hugging Face Model Authentication** card in the Resource Check step tells you when that applies. Enter your token in the field on that card, and the check reruns immediately.
See [Register a Hugging Face account and get an access token](https://docs.ardge.com/en/knowledge/arx-os/huggingface-token) for how to accept the terms and generate the token.
## Step 2: Confirm the model is Running
Go to **Model Instances**. The page lists every deployed instance with its current status:
- **Running** — the model has started and can serve apps or the API
- **Error** — the model didn't start. Open the **Logs** action to read the instance's full log
- Any in-progress status — the system is still starting, stopping or adjusting the model
You can deploy the same model more than once. Each instance serves requests independently, which is how you scale throughput for busy applications — you don't need more than one to start.
## Step 3: Let an app use the model
Two ways to use the model:
- **An app from App Market** — install it and pick your model instance in its settings
- **Another system in your company, or an external app** — point it at the ARX model endpoint
For an application outside the device, use the API endpoint on the **API Key** page. You'll need:
- The endpoint URL
- An API key
- The model instance name
See [Call on-premises models on ARX through the API](https://docs.ardge.com/en/knowledge/arx-os/api-model-calls) for the request format and working examples.
## Step 4: Watch model and GPU usage
**Dashboard** shows how much each model instance is being used and how much of each GPU card is committed. Check it before you deploy another large model — a GPU card with too little VRAM left will reject the deployment.
See [What does GPU VRAM Allocated mean?](https://docs.ardge.com/en/knowledge/arx-os/gpu-vram-allocated) for how the estimate works.
## Where am I right now?
- You only see a model card → it isn't downloaded yet
- It's in **Downloaded Models** → downloaded, but not necessarily started
- **Model Instances** shows **Running** → the model is ready to use
- An app has picked that instance → the app can start calling the model
## Next
With a model running, head to [App Market](https://docs.ardge.com/en/guide/use-ai/app-market) and install the AI application you want to use.
## Sitemap
See the full [sitemap](https://docs.ardge.com/sitemap.md) for all pages.