# AI Agents

> The three kinds of agent, and where each thing lives

## What an AI Agent is

An **AI Agent** is an assistant that serves your customers on its own, with nobody on the other end. The three kinds Audara can build live in the same place and are put together with the same editor: go into **AI Agents** in the left-hand menu and there is the list of the ones you already have.

This page is the map of the group: which kind of agent you need, what all three share and which article covers each thing.

## Which kind you need

The first thing Audara asks you when you create an agent is which kind it is going to be. The window opens by itself, on top of the editor, and until you pick one you cannot go on.

![The window that opens when you create an agent](https://docs.audara.io/agentes-ia/imagenes/agentes-tipo.jpg)

*The window that opens when you create an agent*

|  | **1.** Chatbot | **2.** Voicebot | **3.** Automation |
| --- | --- | --- | --- |
| **How it talks** | In text: WhatsApp, your website chat, Instagram, Messenger or Telegram. | By voice: it answers the call and talks to whoever dialled. | It does not talk at all. It does the work and finishes. |
| **What starts it** | The customer, when they write to the channel. | The customer, when they call the number. | A trigger: by hand, at a set time, an event inside Audara or an HTTP call. |
| **Where you put it to work** | On the channel, under **Setup > Chat channels**. | On an IVR, which is what connects a number to the agent. | On the agent itself, when you configure its trigger. |
| **Where to go next** | [Chatbot](https://docs.audara.io/en/chatbots/) | [Voicebot](https://docs.audara.io/en/voicebots/) | [Automation](https://docs.audara.io/en/autonomo/) |

> **Important**
> The kind is chosen once and cannot be changed afterwards. The window only appears while the agent still has no kind, and closing it takes you back to the list without having created anything. Inside the editor the kind shows as an icon beside the name, but there is nowhere to change it. If you got it wrong, create another agent.

## What all three share

Whichever one you pick, what you build is a **flow**: a sequence of steps the work moves through. Each step does one thing and decides where it goes next. That editor is the same for all three, and the same tools come with it.

- **Several flows in one agent**: The main one and the ones you create alongside it, to keep what repeats in one place. The picker is in the toolbar, and beside it the menu with the flow options.
- **Variable list**: Everything the agent can read and write while it works, with the exact name to copy into a message.
- **Time condition**: When it is on duty. Outside those hours the agent does not start.
- **Select LLM and Model**: Which AI integration the agent thinks with, and on which model. Both fields are in the configuration window.
- **Draft and published**: What you edit is a draft. What actually serves customers is the last published version, and the toggle in the toolbar tells you which of the two you are looking at.
- **Change history**: Who changed what, and when. And if somebody else is editing the same agent, a notice appears above the canvas.
- **File**: Download the flow and upload it again, to carry it over to another agent or keep it aside.

How all of that is used is in [Chatbot](https://docs.audara.io/en/chatbots/), and the full catalogue of steps in [the steps in a flow](https://docs.audara.io/en/chatbot-pasos/).

## What it does besides talking

An agent does more than send and receive messages. From the flow it can go and fetch data from outside, hand a decision over to the AI and store whatever comes back.

- **Smart functions**: These are the abilities you lend the agent: capturing a value out of what the customer wrote, looking something up in a knowledge base or resolving a task with AI. See [Smart Functions](https://docs.audara.io/en/funciones-inteligentes/).
- **A Smart Agent**: For when you want the AI to decide what to do instead of walking a menu. It is three steps working together. See [Smart Agent](https://docs.audara.io/en/smart-agent/).
- **Tools from an MCP server**: So the agent can use tools that live in another system. See [MCP](https://docs.audara.io/en/mcp/).
- **Services over API**: Calling a service of yours in the middle of the flow and carrying on with what it returns. See [API Rest](https://docs.audara.io/en/integraciones-api-rest/).

## What only the voicebot has

The voicebot carries everything that has to do with the ear, and that is why its configuration window has blocks the other two do not.

- **Voice provider (TTS) and Voice type (TTS)**: Which service the agent speaks with, and in which voice.
- **Voice Recognition (ASR)**: Which service it understands what it is told with.
- **Time conditions**: How long it waits in silence before giving up, how often it repeats the question and whether it can be interrupted while it speaks. Careful with the name: in English this block reads almost the same as the Time condition field above, which is the opening hours and a different thing.
- **Limit**: What it does once the ceiling of voicebots serving at the same time has been reached: which message it plays and where it sends that call.
- **SSML Tags**: To adjust how a sentence sounds. The reference is in the flow menu, and also in [SSML tags](https://docs.audara.io/en/etiquetas-ssml/).

On top of that, the toolbar has a phone button listing the IVRs where it is set to answer, which is the question asked most while you are testing one. All of it, with the good practice of writing for the ear, is in [Voicebot](https://docs.audara.io/en/voicebots/).

## Where the reports are

What the agents did is not looked at in this module but under **Reports > AI Agents**, which has three views: **Chatbot**, **Voicebot** and **Voicebot History**. There you go through it interaction by interaction, with the whole conversation. What each one holds is in [AI Agent reports](https://docs.audara.io/en/reportes-bots/).

Automations are not there, because they do not produce a conversation. Their log lives in the module itself: in the list, the automation's menu has **View executions**, with the date, the trigger, how long it took and how it ended.