> ## Knowledge Base Index
> Fetch the complete knowledge base index at: https://help.textit.com/sitemap.xml
> Use this file to discover available pages before exploring further.
> Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash).

# Call AI Action

Need to automate your flow even more? Try using Artificial Intelligence. Add a Large Language Model (LLM) such as OpenAI's [ChatGPT](https://openai.com/) or [Anthropic](https://www.anthropic.com/) to unlock the power of AI. 

# What Can AI Do in My Flows?

The possibilities are endless. Here are some ideas:

* [Auto-translate ](https://help.textit.com/en/article/auto-translate-1du7l/)your flows into multiple languages. (Option to use TextIt's built-in LLM)
* **Language learning**: Practice conversation, translate, or quiz vocabulary.
* **Customer service**: Handle inquiries, complaints, or refund requests conversationally.
* **Internal assistant**: Help employees find documents, policies, or HR info.
* **Onboarding bot**: Guide new users or team members through a process step-by-step.
* **Sales assistant**: Qualify leads, book calls, or suggest packages.
* **Task automation**: Integrate with APIs to handle bookings, reminders, or daily reports.
# Call AI in a Flow

1. [Add your LLM](https://help.textit.com/en/article/artificial-intelligence-1c63w5j/).

2. **Build your flow**. In the example below, we're collecting customer complaints. ChatGPT assesses how angry the customer seems and returns that to the flow, where we categorize their sentiments and respond accordingly. If they're only **mildly upset** or **annoyed**, we'll [open a ticket](https://help.textit.com/en/article/adding-a-textit-ticketing-service-1x17pkr/) with an agent who can respond to them later on. If they're **angry** or **very angry**, we'll let the agent know they need to respond right away!

![](https://storage.crisp.chat/users/helpdesk/website/-/7/a/9/f/7a9fc4558a9de000/screenshot-2026-04-17-at-13322_elc5re.png)
![](https://storage.crisp.chat/users/helpdesk/website/-/7/a/9/f/7a9fc4558a9de000/screenshot-2026-04-17-at-13432_qft2t3.png)
# A Closer Look

### Call AI Action

This is where you'll compose your command for your LLM. Try to be very clear! 

![](https://storage.crisp.chat/users/helpdesk/website/-/7/a/9/f/7a9fc4558a9de000/screenshot-2025-04-24-at-12464_1gw4z5d.png =500x384)
Note that we can reference the result of this Call AI command later in the flow using `@locals_llm_output` like in the Split by Expression seen below.

### Split by Expression

Here, you're evaluating the numerical response you asked the LLM to give you based on the customer's input. 

![](https://storage.crisp.chat/users/helpdesk/website/-/7/a/9/f/7a9fc4558a9de000/screenshot-2025-04-24-at-12500_118kw1q.png =500x472)
### Open Ticket

In this example flow, we want to escalate complaints to a human agent. We'll do that by [opening a ticket](https://help.textit.com/en/article/adding-a-textit-ticketing-service-1x17pkr/). As seen in the flow editor above, we've split the sentiments into two buckets: Needs Response and URGENT. These are topics we've already created on the Tickets page. Here's how we've set up the Open Ticket action for **angry** and **very angry**  customers:

![](https://storage.crisp.chat/users/helpdesk/website/-/7/a/9/f/7a9fc4558a9de000/screenshot-2025-04-24-at-12572_1jckn4z.png =500x275)
We've included some details for the agent, including the text of the complaint submitted by the customer that we collected at the beginning of the flow.  


Need more help? Shoot us a message via the support widget at the bottom right corner.