> 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/audio-native-vs-text-native.md).

# Audio native vs Text native

Audio native for voice calls, Text native for chat conversations: how they differ, when to use which, and how to switch.

A realtime agent works in one of two modes. Both use the same building blocks – the **Identity**, **skills** and **MCP tools** – but they handle the conversation differently.

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2F5kdBC12X8166RxDlihgM%2Frealtime-modes-diagram.png?alt=media" alt="Diagram: both modes share Identity, skills and MCP tools. Audio native: caller, speech-to-speech model that hears and speaks in one step, caller; goes live on a phone number. Text native: customer, optional speech to text, language model that writes the answer, optional voice, customer; goes live in chat bubbles"><figcaption><p>How Audio native and Text native handle a conversation</p></figcaption></figure>

## The two modes

|                   | Audio native                                   | Text native                                                                |
| ----------------- | ---------------------------------------------- | -------------------------------------------------------------------------- |
| **Model hears**   | The caller's voice, directly                   | Text – a transcript when speech-to-text is on                              |
| **Agent replies** | Speaks in one step, with the model's own voice | Writes text; a separate voice can read it                                  |
| **Models**        | Realtime (speech-to-speech) models only        | Any language model – the same models as classic flows                      |
| **Speed**         | Fastest, most natural turn-taking              | Slower when speech is on, because the steps add up                         |
| **Goes live on**  | A phone number                                 | Chat bubbles (phone numbers are not available yet)                         |
| **Settings**      | Model & Voice, Listening, Transcript           | Provider settings (LLM, Speech to text, Voice), Response settings, Timeout |

### Audio native

{% embed url="<https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2FExfyafIH9WUHdOWBDv23%2Faudio-native-explainer-v2.mp4?alt=media>" %}
Audio native in 43 seconds: speech-to-speech models, natural barge-ins and quicker replies, and how to build, test and go live.
{% endembed %}

New speech-to-speech models hear the caller and answer with their own voice in one step. There is no separate speech-to-text and text-to-speech, so replies come quicker and the conversation sounds natural. With **Caller can interrupt** on, callers can barge in at any moment – the agent stops, listens and carries on. See [Voice and listening](https://docs.borndigital.ai/realtime-agent/voice-and-listening.md).

### Text native

{% embed url="<https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2FSfX07JUbtBfEzDirUHKU%2Ftext-native-explainer-v3.mp4?alt=media>" %}
Text native in 48 seconds: the new way to build agentic chatbot conversations, built on the realtime principle – when to use it, and how to set it up, test it and go live.
{% endembed %}

Text native is the new way to build agentic chatbot conversations, built on the realtime principle. The answers are written by text-based language models – the same ones as in classic flows – so you can expect the same reliable performance. Speech-to-text and a voice can be switched on, so customers can also talk to the agent in the chat bubble.

## When to use which

* **Audio native** – voice calls where natural turn-taking matters: service lines, information and simple transactions on a phone number.
* **Text native** – chat conversations, for example everything that today runs on **one AI node and a knowledge base**.
* **Flows with tool calls and several branches** – move them into **skills** and tools on an **MCP server**; both modes can use them. See [Prompt and skills](https://docs.borndigital.ai/realtime-agent/prompt-and-skills.md) and [MCP tools](https://docs.borndigital.ai/realtime-agent/mcp-tools.md).

{% hint style="warning" %}
Realtime agents are not yet suitable for conversations that need **deterministic decisions** with Decision nodes. Keep those in a classic flow for now.
{% endhint %}

## Choose or switch the mode

Choose the mode when you create the project: **Audio native** or **Text native** in **Create project**.

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2FBICRLadTYlfYAMBIXwMX%2Frealtime-create-text-native.png?alt=media" alt="The Create project dialog of a realtime agent with the name Docs – Chat assistant and Text native selected instead of Audio native"><figcaption><p>Choosing Text native when you create the project</p></figcaption></figure>

To switch an existing agent, use the **Audio native** / **Text native** tabs at the top of the settings column. Studio asks **Switch to Text native?** (or **Switch to Audio native?**) and compares the two modes. Each mode keeps its own settings – switch back and they return as you left them.

<figure><img src="https://4261467870-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6iQTvxgRZRwPS1NgIGEb%2Fuploads%2FTJGDhf3phPky1KoDmNDv%2Frealtime-test-agent-mode.png?alt=media" alt="The dialog Switch to Text native? comparing Audio native (now) and Text native: what the model hears, how the agent replies, speed and models, with Cancel and Switch to Text native"><figcaption><p>Switching an agent to Text native</p></figcaption></figure>

Text native has its own settings: **Provider settings** with the tabs **LLM** (provider and model), **Speech to text** and **Voice**, **Response settings** (reasoning effort and summary for supported models) and **Timeout** (how long to wait for a response before giving up). How to test each mode and go live: [Test and go live](https://docs.borndigital.ai/realtime-agent/test-and-go-live.md).
