> ## 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.

# Inworld

> Connect Inworld to SentiVue Talk

Inworld is a speech platform whose text-to-speech models are built for low-latency conversation. In SentiVue Talk it provides the **voice** that speaks your agent's responses.

## What you need

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

## Getting your API key

1. Sign in to the [Inworld Portal](https://platform.inworld.ai/).
2. Click **API Keys** in the bottom left sidebar.
3. Click **Generate new key**.
4. Copy the **Basic (Base64)** authorization signature.

<Warning>
  Step 4 is the one people get wrong. The portal shows a key and a secret alongside the **Basic (Base64)** signature, and only the Base64 signature is what SentiVue Talk accepts. Copy that value — not the raw key, not the secret, and not the two joined together.
</Warning>

Keys come in types, and the type is fixed when the key is created. Generate a **Standard** key. A **Realtime-only** key is scoped to Inworld's Realtime API and will not drive text-to-speech.

Keys belong to the workspace that was selected when you generated them, and a cloned or designed voice belongs to one workspace. Generate the key in the same workspace as the voices you intend to use.

## Adding the connection

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

SentiVue Talk validates the key with Inworld 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 model and voice on [Model Configurations](/talk/configurations/inference-providers).

**Model** — start with the flash variant on calls and move up only if the voice quality falls short on a real call. See Inworld's [TTS models page](https://docs.inworld.ai/tts/tts-models) for the current line-up.

**Voice** — a voice is referenced by name, for example `Dennis`. Browse the stock, community and your own voices in the [Voice Library](https://platform.inworld.ai/voice-library), or audition them in the [TTS Playground](https://platform.inworld.ai/tts-playground).

**Custom voices** — a cloned or designed voice is referenced by its own resource name; one published to the community library uses its public `community-` identifier.

**Languages** — BCP-47 codes, from broad codes like `es` through regional locales like `es-MX`. Some are marked experimental and vary more in pronunciation — check yours on Inworld's [language support page](https://docs.inworld.ai/tts/capabilities/multilingual).

## Completing the pipeline

Inworld drives voice only. You also need:

* A [language model](/talk/configurations/llm) to decide what the agent says
* A [transcriber](/talk/configurations/transcriber) to turn caller speech into text

Each is a separate [connection](/talk/configurations/connections).

## Troubleshooting

<AccordionGroup>
  <Accordion title="The key is rejected when I save">
    You most likely copied the raw key or the secret rather than the **Basic (Base64)** authorization signature. Return to **API Keys** in the Inworld Portal and copy that value. Also confirm the key is a Standard key — a Realtime-only key cannot drive text-to-speech, and a key's type cannot be changed after it is created.
  </Accordion>

  <Accordion title="A voice I can hear in the portal is not found on a call">
    Cloned and designed voices belong to a single Inworld workspace, and a key only reaches the workspace it was generated in. Switch to that workspace in the portal, generate a key there, and update the connection. For a community voice, use its public `community-` identifier rather than the display name.
  </Accordion>

  <Accordion title="Calls worked and then stopped mid-month">
    Inworld bills against a monthly credit balance shared across every product on the account, and text-to-speech is charged per character synthesized. Check the balance on the portal's Billing page and top it up or enable auto-reload. A failed subscription payment also downgrades the account and disables API usage.
  </Accordion>

  <Accordion title="Some calls lose the voice under load">
    Inworld caps how many generations run at once per account, shared across all your API keys, and requests over the cap are rejected. The cap is set by your plan. If you run many simultaneous calls, raise the plan.
  </Accordion>
</AccordionGroup>