> For the complete documentation index, see [llms.txt](https://docs.borndigital.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.borndigital.ai/realtime-agent/introduction.md).

# Introduction

Voice agents that talk to callers in real time with a speech-to-speech model: what they are, how they work and how the editor is organised.

A **realtime agent** is a voice agent that answers phone calls with a **speech-to-speech model**, such as OpenAI's gpt-realtime or Gemini Live. The model listens to the caller and answers with its own voice directly – there is no conversation flow with nodes. You describe the agent's role and rules in a prompt, connect the systems it may use as MCP tools, and put it on a phone number.

## Realtime agent or Digital Agent?

|                    | Realtime agent                                                         | Digital Agent                                              |
| ------------------ | ---------------------------------------------------------------------- | ---------------------------------------------------------- |
| How you build it   | A prompt, optional skills and MCP tools                                | A conversation flow of nodes                               |
| Channels           | Phone calls                                                            | Chat bubble and phone calls                                |
| Conversation style | Natural and fast; the caller can interrupt                             | Exactly as designed; every step under your control         |
| Best for           | Open conversations, service lines, information and simple transactions | Strict processes, forms, identification, regulated scripts |
| Handing over       | Can end the call; no transfer to an operator                           | Transfer and redirect to an operator                       |

## How it works

1. A customer calls a phone number the agent is deployed to.
2. The agent greets the caller – the greeting is generated from your prompt.
3. The speech-to-speech model listens and answers. It follows your **prompt**, switches to a **skill** when the conversation needs it, and calls **MCP tools** to look up or change data in your systems.
4. The conversation is transcribed. After the call, the recording or transcript can be sent to an [Insights](/insights/introduction.md) project for analysis.

## Open the Realtime product

After signing in, choose **Realtime** on the product screen (or in the **Select product** menu in the sidebar). You see **All projects** of your organization – see [Projects](/realtime-agent/projects.md).

## The editor

When you open a project, the editor shows:

* **Left:** **Agent** and the list of skills, with **New Skill**.
* **Centre:** the agent's **Identity** (its prompt) and **MCP servers** – or the selected skill – and a graph of the agent and its skills.
* **Right:** the agent name and its deployed version (`v1`, `v2`, …), **Settings**, **History**, undo and redo, and below them the settings of the voice, transcript, listening and phone number, with **Go live**.
* **Bottom right:** the **Voice chat** bubble for test calls in the browser.

The left navigation also leads to **Analytics** – the connection to Insights.

Several people can edit the same agent at the same time; their changes appear almost immediately.

## Build your first realtime agent

1. [Create a project](/realtime-agent/projects.md#create-a-project).
2. Write the [prompt](/realtime-agent/prompt-and-skills.md).
3. Optionally connect [MCP tools](/realtime-agent/mcp-tools.md).
4. Choose the model, voice and listening behaviour in [Voice and listening](/realtime-agent/voice-and-listening.md).
5. [Test it in the browser and go live](/realtime-agent/test-and-go-live.md) on a phone number.
6. Optionally send the calls to [Insights](/realtime-agent/analytics-with-insights.md).