# Connect an Azure OpenAI resource

> Azure asks for two fields rather than one - a resource endpoint and a key - and lists nothing until you have deployed a model yourself.

- Canonical: https://big-agi.com/docs/connect-azure
- Minimum tier: open  [Open ⊂ Free ⊂ Pro - a tier only ever ADDS to the Open baseline]
- Kind: provider · Last content update generation: 2026-07-27 · Review by: 2026-10-27
- Answers: Where do I get the endpoint and the key? · Why do I have to create a deployment before any model appears? · Does my company's Azure subscription cover this?

Azure asks for two fields: a resource endpoint and a key, both issued by your own cloud subscription. Big-AGI supports the classic Azure OpenAI Service resource shape. The newer Foundry resource enumerates models differently and is not supported yet.

## The subscription behind it

An [Azure subscription](https://azure.microsoft.com/pricing/purchase-options/azure-account) is the prerequisite, usually one you or your company already pay for. Usage lands on the Azure invoice. The enforceable limit is an Azure Cost Management budget on that subscription.

## Creating the resource and deploying a model

1. Create the resource from the [portal](https://portal.azure.com/#create/Microsoft.CognitiveServicesOpenAI). Its Basics tab asks for **Subscription**, **Resource group**, **Region**, **Name** and **Pricing Tier** ([Microsoft's own steps](https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/create-resource)). Region decides which models you can deploy, and changing it later means a new resource. Check [availability by region](https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure-region-availability) first. Opinion, July 2026: East US, West Europe and Australia East carry the widest selection.
2. On the **Network** tab take the first option, `All networks, including the internet, can access this resource.` - the default. The other two put the resource behind a virtual network or a private endpoint. Nothing reaches it from your browser or from the Big-AGI fast edge servers. The failure reads as a bad key.
3. Deploy at least one model at [ai.azure.com](https://ai.azure.com/). Nothing is callable before this happens.
4. Read the endpoint and `KEY 1` off the resource blade.

The Basics tab of the Create form asks for the five fields above, resource group and region first.

![The Azure portal Create Azure OpenAI form, Basics tab, with resource group and region highlighted](/docs/legacy/config-azure-openai-create.png)

## The two fields, and what fills the list

| Field in Big-AGI | What to paste |
|---|---|
| **Azure Endpoint** | The resource endpoint, `https://<resource>.openai.azure.com` |
| **Azure Key** | `KEY 1` from the resource's keys page |

What arrives in the model list is your deployments, under the names you gave them - the deployment name is the identifier, not the model name. Deployments belong to a region, so a resource in one region shows nothing you deployed in another.

## Two API paths behind one endpoint

| Path | When Big-AGI takes it | What it calls |
|---|---|---|
| v1 | GPT-5-class models, on by default | `/openai/v1/responses`, no deployment id |
| Deployment-based | Older models, and whenever v1 is off | `/openai/deployments/{deployment}/{function}` |

The deployment name is therefore the identifier on the deployment-based path only. Listing is unaffected: the model list always comes from the deployments API. A server you operate forces the older path with `AZURE_OPENAI_DISABLE_V1=true` ([keys held by a server you operate](/docs/self-host-configuration)).

## Filling both fields in Big-AGI

`Ctrl + Shift + M` opens **Models**; press **More Services** if the **Setup AI Models** wizard is in front. Then **Add** -> **Azure OpenAI**, fill both fields, and press **Models** ([add and manage your keys](/docs/connect-models)).

## Direct Connection needs both fields

Direct Connection can be turned on once both the endpoint and the key are in the browser. It starts off.

> **Direct Connection works only with your API key stored in the browser**, and with an AI service that permits direct browser calls (CORS). Where it cannot be used, requests route through the Big-AGI fast edge servers instead - everything still works, within the standard upload size and time limits.

## When this doesn't work

| What you see | What it means | What to do |
|---|---|---|
| `Azure OpenAI API Host is invalid:` | The endpoint is not a well-formed URL. | Re-copy the endpoint from the resource blade |
| Web search absent on a model that has it everywhere else | Azure carries no `web_search_preview` tool, as of 2025-11-18, so Big-AGI removes the parameters that enable search. Nothing on screen says so. | Run that model on another AI service when the answer needs live sources |

## Related

- [Connect an AI service](/docs/connect-models)
- [Keys held by a server you operate](/docs/self-host-configuration)
- [Connect any OpenAI-compatible endpoint](/docs/connect-custom-endpoints)