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

> ## Agent Instructions
> Responda sempre em português do Brasil, com os nomes de menus e botões exatamente como aparecem no painel do Notfy.
> Use somente o que está documentado aqui. Se a documentação não cobrir a pergunta, diga isso e indique a página mais próxima em vez de inventar.
> Funcionalidades podem depender do plano, do segmento da empresa, das permissões do usuário e de integrações ativas: cite essa condição quando a página a mencionar.
> Nunca peça nem repita senhas, tokens ou dados pessoais de clientes.

# AI agent in customer service

> Turn on the AI agent that answers leads on WhatsApp and Instagram, and learn what it does, what it does not do and how it works in each segment.

The AI agent answers customer messages in the chat using the name, context and rules you defined. It collects lead data, checks the calendar and schedules meetings, and calls the team when needed.

## When to use

* Serve leads outside team hours or during peak times.
* Make the first contact, qualify the lead and leave the appointment already booked.
* Reduce response time without growing the team.

## Before you start

* Have a connected [official WhatsApp](https://docs.notfy.ai/en/configuracoes/integracao-whatsapp-oficial.md) or [Instagram](https://docs.notfy.ai/en/configuracoes/integracao-instagram.md) integration.
* Have AI credits available. See [AI credits](https://docs.notfy.ai/en/ia/creditos.md).
* Have filled in the [knowledge base](https://docs.notfy.ai/en/ia/base-conhecimento.md).
* Have permission to edit integrations (administrators).

> [!WARNING]
>
> When you turn on AI, the screen shows a notice that the feature uses artificial intelligence with processing in the United States and that, by enabling it, you agree to the international transfer of data (LGPD, art. 33). Read it before turning it on.

## How to turn on AI on a channel

AI is turned on per integration, which means you choose which numbers and accounts it serves.

> [tab] WhatsApp
>
> **1. Open the integration**
>
> Go to **Settings**, **Integrations** and open the WhatsApp integration.
>
> **2. Go to the WhatsApp tab**
>
> Open the **WhatsApp (Official Meta)** tab and find the **Official Behavior** section.
>
> **3. Turn on AI**
>
> Enable **Enable AI on this number**.
>
> **4. Decide on automatic leads**
>
> If you want, also enable **Automatic Lead Creation**, so that new contacts become leads in the pipeline.
>
> **5. Save**
>
> Save the integration.

> [tab] Instagram
>
> **1. Open the integration**
>
> Go to **Settings**, **Integrations** and open the Instagram integration.
>
> **2. Go to the Instagram tab**
>
> Open the **Instagram (Official Meta)** tab and find the **Instagram Behavior** section.
>
> **3. Turn on AI**
>
> Enable **Enable AI on Instagram**. This option only appears when **Receive Direct messages (Chat)** is on.
>
> **4. Save**
>
> Save the integration.

## One AI profile for all channels or one per channel

In **Company settings**, **AI** tab, the **Apply this setting to all channels** option defines how the agent behaves.

| Option | Effect |
| :- | :- |
| On (default) | The company's name, context and rules apply to the AI on all channels. |
| Off | Each integration has its own name, objective, context and rules, in the integration's **Artificial Intelligence** tab. |

With the option off, the integration also gets the **Service Mode** field:

* **Artificial Intelligence**: the agent leads the conversation.
* **Fixed Flow (visual builder, no AI)**: a flow built by you, with template sending, waiting for the reply and branches, which replaces the AI on that channel. The **Open Flow Editor** button takes you to the builder.

## What the agent does

* Replies using the name you gave it (**AI Name** field).
* Follows the objective chosen in **Main AI Objective** (see the table below).
* Collects lead data during the conversation, such as name and contact.
* Checks free slots in the calendar and creates the appointment. If Google Calendar is connected, the event is also sent to it.
* Understands audio messages: the audio is transcribed and answered as text.
* Hands the conversation over to the team when needed. See [Hand over to an agent](https://docs.notfy.ai/en/ia/atendimento-humano.md).

### Available objectives

| Objective | Recommended for |
| :- | :- |
| Sales consulting (consortia) | Consultative sales with simulation scheduling. |
| Legal consulting (law firms) | Case screening and consultation scheduling. |
| Vehicle sales (car dealership) | Service about vehicles in stock and test-drive scheduling. |
| Customer service (any niche) | Generic service, without assuming a specific industry. |

At a car dealership, the agent also searches real vehicles in your stock and records the lead's interest. The AI never invents vehicles.

## What the agent does not do

* It does not answer conversations that are already with a human agent.
* It does not answer when AI credits run out. The conversation stays waiting for human service.
* It does not confirm a meeting date or time that the customer has not said and that the calendar has not released.
* It does not send emojis and asks one question at a time.
* It does not decide on its own about formal complaints, billing disputes or complex legal matters: in those cases it calls the team.

## Business hours

> [!NOTE]
>
> There is no schedule to turn the AI on and off: it answers at any time, as long as there are credits and the conversation is not with a human.

Hours only matter when scheduling. To suggest free slots, the agent uses the business hours and lunch break written in the **About the company** field. If nothing is stated, it assumes 9 a.m. to 6 p.m. For that reason, describe the company's service hours in the knowledge base.

For the interval between meetings and the default duration, see the **Interval between meetings** and **Meeting duration** fields in [General settings](https://docs.notfy.ai/en/configuracoes/parametros-gerais.md).

## Response time

In **General Settings**, **AI** tab, the **Response Delays** section defines the **Minimum time** and **Maximum time** to wait. The AI picks a random value between the two before replying, so the conversation feels natural. Long answers are split into short messages, sent with small intervals.

## Frequently asked questions

> [accordion] The AI replied and then stopped. Why?
>
> The most common reasons are: the conversation was taken over by an agent, the AI called the team, the credits ran out or the conversation was finished. Check the conversation mode in the details panel of the [Chat](https://docs.notfy.ai/en/atendimento/chat.md).

> [accordion] Can I turn on the AI on just one number?
>
> Yes. The option is per integration: turn it on only in the integrations where you want the agent.

> [accordion] How do I fix a wrong AI answer?
>
> Adjust the text in **About the company** and **Service rules**. The more specific the rule, the better the result. If the information is not in the base, the AI should not make it up.

> [accordion] What happens if the AI is unavailable?
>
> If there are consecutive failures, the AI is paused for a few minutes for that company and the conversation is not answered automatically. During that period, follow the **Queue** tab of the chat.

> [card] [Knowledge base](https://docs.notfy.ai/en/ia/base-conhecimento.md)
>
> Teach the AI about your business.

> [card] [Hand over to an agent](https://docs.notfy.ai/en/ia/atendimento-humano.md)
>
> When and how the team takes over.

> [card] [AI credits](https://docs.notfy.ai/en/ia/creditos.md)
>
> Understand consumption.

> [card] [Multichannel chat](https://docs.notfy.ai/en/atendimento/chat.md)
>
> Follow the conversations.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.