> ## 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.

# OpenAI

> Connect OpenAI to SentiVue Talk

OpenAI covers three parts of the pipeline: the **language model** that decides what the agent says, the **voice** that speaks it, and the **transcription** that turns the caller's speech into text.

One connection drives all three. You do not add OpenAI once per service.

## What you need

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

## Getting your API key

1. Sign in at [platform.openai.com/api-keys](https://platform.openai.com/api-keys).
2. Create a new secret key.
3. Copy it immediately — OpenAI shows the key once and never again.

Keys begin with `sk-`.

<Note>
  OpenAI keys belong to a **project**, with its own billing, rate limits and usage tracking. Create the key in the project you want the usage charged to, and make sure that project has billing set up — a key with no payment method behind it authenticates but fails on the first call.
</Note>

For anything running in production, use a key from a **service account** rather than one created under a personal login. Personal keys are revoked when their owner loses access to the project, which stops calls with nothing on the SentiVue Talk side to explain it.

## Adding the connection

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

SentiVue Talk validates the key with OpenAI as you save, so a bad or expired key is caught immediately rather than on a live call.

## What you can select afterwards

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

**Language model** — pick from the dropdown, or type a name in for any model your account can reach. The `mini` and `nano` variants are faster, and a slow reply is heard as a pause. See OpenAI's [model list](https://platform.openai.com/docs/models).

**Voice** — pick from the dropdown. OpenAI's [text-to-speech page](https://platform.openai.com/docs/guides/text-to-speech) has samples.

**Language** — set it to match your callers. See OpenAI's [speech-to-text guide](https://platform.openai.com/docs/guides/speech-to-text).

## Using OpenAI for only some services

A connection can be limited to specific services. If you want OpenAI for the language model but a different provider for the voice, turn the voice service off on the OpenAI connection.

## Custom hosts

If you reach OpenAI through a proxy or gateway rather than `api.openai.com`, you can set a base URL per service on the connection. Each service can point somewhere different.

## Realtime is a separate connection

OpenAI's speech-to-speech model is a separate provider in SentiVue Talk, with its own connection. See [OpenAI Realtime](/talk/configurations/providers/openai-realtime).

## Troubleshooting

<AccordionGroup>
  <Accordion title="The key is rejected when I save">
    Confirm you copied the whole key including the `sk-` prefix, that it has not been revoked, and that the project it belongs to has billing enabled.
  </Accordion>

  <Accordion title="Calls fail with a quota or rate-limit error">
    Rate limits are per project, not per organization. Check the limits on the project the key belongs to rather than your account total.
  </Accordion>

  <Accordion title="The key worked for months and then stopped">
    If it was created under someone's personal login, check whether that person still has access to the project. Personal keys are revoked along with their owner's account. Replace it with a service-account key.
  </Accordion>
</AccordionGroup>