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

# Voice Interactions

> Let your users practice a conversation out loud with an agent that plays a role and speaks back.

A Voice Interaction is a spoken roleplay you set up in the Agent Management Studio. Someone talks to the agent the way they would talk to a real person, and the agent replies out loud in a voice you choose, staying in a role you define, such as an unhappy customer, a job interviewer, or a coach. You describe the role and the situation in plain English, pick a voice, and assign it to an agent. Your users then start the conversation from the chat and speak to it directly.

You build a Voice Interaction once and reuse it. Each one is saved and versioned, so you can keep improving it, and any agent can use it, either by offering it in conversation or by running it as a step in a Plan.

## Demo

<iframe className="w-full aspect-video rounded-xl" src="https://www.loom.com/embed/874be69831d64553b1692b790198f099" frameBorder="0" allowFullScreen />

## Key features

### A two-way spoken conversation

Your user speaks, and the agent replies out loud in real time. It works like a phone call.

### Set it up by describing it

Open the Interaction Builder, tell it who the agent should play and what the situation is, and it writes the setup for you. Carry on chatting to adjust anything you want to change.

### Your own voice provider

Voice Interactions run on ElevenLabs or Deepgram using your own account and API key. Connect more than one, and choose the provider and voice separately for each interaction, so a calm coaching scenario and an angry-customer scenario can sound completely different.

### Saved and versioned

Every interaction is saved, named, and versioned. You can improve one over time without losing earlier versions, preview any version, and set the one you want to be live.

### Use it across your agents and Plans

Assign an interaction to any agent. The agent can offer it in conversation, show it as a welcome card at the top of the chat, or run it at a set point in a Plan. The same interaction works for your own team and for the users of an agent you have embedded in your product.

## Use cases for your end users

* A manager rehearses giving difficult feedback before the real conversation.
* A candidate runs a mock interview and gets coached on their answers.
* A sales or support rep practices a tough call, such as an angry customer or a hard objection.
* A trainer builds a roleplay once and reuses it across a whole cohort.
* A customer embeds an agent in their own product and offers the same practice to their users, not just to internal staff.

## Current limitations

* Each interaction uses one standard voice card, and it takes on the look of the chat it sits in.
* Usage and cost are tracked on your provider's dashboard, not inside Mindset AI.
* Voice Interaction sessions have a default limit of 10 minutes.
* English language support only.

## Getting started

<Steps>
  <Step title="Connect a voice provider">
    In AMS, open Voice Interactions and go to Settings to connect your ElevenLabs or Deepgram key.
  </Step>

  <Step title="Build your first interaction">
    Follow the [Create a Voice Interaction](/build-optimize/AMS/create-a-voice-interaction) guide to build and publish your first one.
  </Step>
</Steps>