# AI actions in BRIX AI

> [HTML Version](cortex-ai-actions.html)

AI actions in **BRIX AI** allow you to embed artificial intelligence capabilities into business processes, interfaces, and scripts in BRIX. An AI action automatically performs specified steps based on input data and returns the result.

For example, you can add an AI action to a page to quickly summarize customer emails. An employee pastes the email text into a field, clicks a button, and gets a concise summary generated in a separate field.

AI actions also support JSON data transfer, passing information as a structured set of fields. Suppose you add an AI action to a business process to classify customer requests. Request details are passed from the process context, and the AI action saves a text analysis and the identified request type, such as issue or suggestion, directly to process variables.

**Important**: This article describes features available in the **BRIX AI Enterprise** edition. Some tools may not be available in other editions. To learn more, see [Available BRIX AI editions](cortex-editions.md).

## How AI actions work

Here is how AI actions are configured and used:

1. Create an AI action in **BRIX AI**.

To define its behavior and capabilities, configure the following settings:

- Language model prompt. Enter a text description of the task and expected output for the AI action.

- Input data format and response schema. Choose whether to pass data as plain text or as a JSON object. Plain text works well for basic tasks, while JSON makes data parsing and model output predictable.

- Tools: For tasks beyond the model's built-in capabilities, add tools to your AI action. Tools define where the action gets information, what data it processes, and what operations it carries out in the system. You can connect:

	- Prebuilt tools to work with date and time, files, images, external web content, or data from BRIX.

	- Custom BRIX tools that allow the AI action to interact with apps, processes, and tasks.

	- Smart search projects that enable searching across configured knowledge bases.

	- Child **BRIX AI** components that use other AI actions or specialized AI agents as tools to deliver precise results. This allows you to [route complex tasks](#task-routing).

	- Custom tools via MCP servers to configure custom capabilities, such as fetching data from external systems.

2. Set up execution triggers in BRIX once the AI action is ready.

You can trigger an AI action from various system components:

- In business processes. Add a dedicated process shape and select the AI action in its settings.

- In the user interface. Add a custom widget to a form or page in BRIX and link it to the AI action. Users can run the action by clicking a button.

- In scripts. Call the AI action from any system script using the API.

## Task routing in AI actions

To improve accuracy, you can split complex tasks across multiple **BRIX AI** components. Configure a parent AI action responsible for the overall task, then create child components for specialized steps. Connect the child components to the parent AI action in one of the following ways:

- As a tool. Connect an AI action or an [AI agent](cortex-ai-agents.md). The child component receives input data from the parent AI action, processes it, and returns the result.

- As a managed component. Connect a child AI agent. The agent receives the full execution context from the parent AI action, allowing it to complete the task with complete contextual awareness.

For instance, a main AI action categorizes customer requests into issues, suggestions, or other types. Depending on the type, it routes the request to dedicated child components optimized for processing that specific category. This task routing ensures higher quality and more accurate outcomes.

To learn more about AI actions, see:

- [Create and configure an AI action](configure-ai-action.md).

- [Debug an AI action](ai-action-calibration.md).

- [Call an AI action in BRIX](call-ai-action.md).