> ## 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.

# Ultravox

> Connect Ultravox to SentiVue Talk

Ultravox is a speech-native voice AI platform whose models drive **realtime speech-to-speech** conversations in SentiVue Talk. The model hears the caller and speaks back over one connection, so no separate transcriber or voice provider sits in the path.

## What you need

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

## Getting your API key

1. Sign in at [app.ultravox.ai](https://app.ultravox.ai/).
2. Open [**Settings**](https://app.ultravox.ai/settings/).
3. In the API Keys section, choose **Generate New Key** and name it.
4. Copy the key and store it somewhere safe before leaving the page.

## Adding the connection

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

A wrong or expired key is not caught at this point — it first shows up on a live call. Place a test call after saving to confirm the connection works.

## What you can select afterwards

Turn on **Realtime** in [Model Configurations](/talk/configurations/inference-providers) and choose Ultravox as the provider.

**Model** — pick a suggestion or type any model name Ultravox accepts; the default is `ultravox-v0.7`. See Ultravox's [models endpoint](https://docs.ultravox.ai/api-reference/other/models-get) for what your account can use.

**Voice** — a free-text field with no dropdown, so you enter either the voice name or its voice ID rather than picking from a list. Find both in Ultravox's [voices explorer](https://app.ultravox.ai/voices).

**Language** — Ultravox covers 26 spoken languages. Voices are tied to a primary language, so choose one that matches the language your agent speaks.

## How a speech-to-speech call differs

A speech-to-speech model replaces the transcriber and the voice provider. When a workflow is set to realtime, those two settings no longer apply.

* **The voice comes from the model.** You cannot pair Ultravox with a voice from another provider.
* **You still need a language model configured.** It runs the work the realtime model does not — extracting variables and running QA analysis after the call. Without one, those features silently produce nothing.
* **Context compaction does not apply** and is ignored if enabled.

## Troubleshooting

<AccordionGroup>
  <Accordion title="The connection saved fine but calls fail immediately">
    Saving does not check the key against Ultravox. Confirm the key is still listed and active under Settings in the Ultravox console, and that you pasted it whole.
  </Accordion>

  <Accordion title="The agent connects but uses the wrong voice, or the call drops at the start">
    The voice field is free text and is not checked when you save it. Confirm the name or voice ID matches an entry in the [voices explorer](https://app.ultravox.ai/voices) exactly.
  </Accordion>

  <Accordion title="Calls worked, then started failing partway through the month">
    A new Ultravox account includes a limited allowance of free call minutes. Once it is used up, calls need billing set up on the account. Check the account's usage and payment method in the Ultravox console.
  </Accordion>

  <Accordion title="Calls fail once several run at the same time">
    Pay-as-you-go Ultravox accounts are capped at 5 concurrent calls, and anything above that is rejected outright. Upgrade the plan if a campaign needs more, or lower the workflow's concurrency.
  </Accordion>

  <Accordion title="Variables are not extracted and QA never runs">
    Both need an LLM configured alongside the realtime provider. Check the LLM tab in Model Configurations.
  </Accordion>
</AccordionGroup>