> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sentivue.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Azure OpenAI

> Connect Azure OpenAI to SentiVue Talk

Azure OpenAI serves OpenAI's models from your own Azure resource, under your subscription's billing and region. In SentiVue Talk it drives the **language model**.

## What you need

| Field    | Required |
| -------- | -------- |
| API key  | Yes      |
| Endpoint | Yes      |

The endpoint is your resource's own hostname and has no default.

## Getting your credentials

1. Sign in to the [Azure portal](https://portal.azure.com/#browse/Microsoft.CognitiveServices%2Faccounts/kind/OpenAI) and open your Azure OpenAI resource.
2. Go to **Keys and Endpoint**.
3. Copy the **endpoint**. It is the hostname of your resource, in the form `https://YOUR-RESOURCE-NAME.openai.azure.com/`. Every resource has its own.
4. Copy **KEY 1** or **KEY 2**. Azure issues two keys per resource so you can rotate one while the other stays live — either works.
   Under **Model deployments**, note the name of each deployment you intend to use. You will need those names later, and they are not necessarily the model names.

## Adding the connection

1. Go to **Connections** and choose **Azure OpenAI**.
2. Paste the API key and the endpoint.
3. Save.

Place a test call after connecting. A wrong key, endpoint, or deployment name saves without complaint and surfaces on the first call.

## What you can select afterwards

Choose your models on [Model Configurations](/talk/configurations/inference-providers).

**Language model** — enter the **deployment name**, not the model name. If you deployed GPT-4o as `my-gpt4o`, enter `my-gpt4o`. See Microsoft's [models page](https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models) for what your region offers.

## Completing the pipeline

This connection has no voice and no transcription. You also need:

* A [voice provider](/talk/configurations/voice) to speak the responses
* A [transcriber](/talk/configurations/transcriber) to turn caller speech into text

[Azure Speech](/talk/configurations/providers/azure-speech) covers both from Azure, though it is a separate connection with its own credential and takes a region rather than an endpoint. Any other provider works just as well. For speech-to-speech, use [OpenAI Realtime](/talk/configurations/providers/openai-realtime).

## Troubleshooting

<AccordionGroup>
  <Accordion title="A model I entered is not found">
    Azure resolves the name against your deployments, not against OpenAI's catalogue. Open **Model deployments** on the resource and copy the deployment name exactly as it appears there. A model that exists in Azure but has not been deployed on your resource fails the same way — deploy it first.
  </Accordion>

  <Accordion title="The endpoint is rejected when I save">
    Use the value from **Keys and Endpoint** as shown, the `https://YOUR-RESOURCE-NAME.openai.azure.com/` hostname. Endpoints are per resource, so a value copied from a different resource or from documentation will not work.
  </Accordion>

  <Accordion title="The key is rejected but I copied it correctly">
    Keys are scoped to one resource. A key from another Azure OpenAI resource authenticates nowhere else, and the pairing with the endpoint has to match. Confirm both were copied from the same resource's **Keys and Endpoint** page. If the key was recently regenerated in the portal, the old value stops working immediately.
  </Accordion>

  <Accordion title="A model available in Azure is not available to me">
    Model availability varies by region, and a model is only reachable once it is deployed on your resource. Check your resource's region against Microsoft's [models page](https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/models), then create the deployment and use its name.
  </Accordion>
</AccordionGroup>