The AI ​​Agent creates generative bots, using predefined templates and designing flows in a simple way. In this way, it facilitates the creation of more efficient experiences for users. With the AI ​​Agent, it is not necessary to create a text block for each interaction, but the AI ​​Agent will interact with the user through text until it meets its objective.

Within the AI ​​Agent block is the space to write the instruction (prompt) for Generative Artificial Intelligence.

To complete the AI Agent block, follow the steps inside.

1. Role and responsibilities. In this step, you must define the agent’s role in under 100 characters and list its responsibilities (without detailing the instruction).

2. Tone and style. In this step of the block, you must define the agent’s response tone and style, complementing the settings made in the bot’s general “Identity and preferences” section.

3. Instructions: In this step of the block, you must detail the step-by-step instruction. This is where you can include variables, actions, or content bases, if applicable.

Here are some tips to help you provide an effective instruction in this step:

https://help.botmaker.com/en/help/6634855130683497129

4. Do not talk about. In this step of the block, the user must define the safety controls that prevent each agent from giving inappropriate responses.

5. Generative model. In this step of the block, the user can choose which model to use for a specific agent. We recommend the following article to help you learn more about this setting.

How to add variables, actions, content bases, and code actions?

To add these elements, simply insert the “$” symbol in the instruction block, as shown in the image below.

The AI agent can interact with the following elements:

• Variables: The agent can access user information stored in variables and use it to fulfill its goal. It can also capture data from user messages and save it in variables if requested in the instructions.
• Actions: The agent has a list of actions it can use to achieve its goal. More actions will be added in the future to further enhance the agent.
• Content bases: The agent can consult content bases to fulfill its goal by searching for information or validating data within them. To learn more about content bases, you can read this article.
• Code actions: The agent can invoke code actions as part of its objectives. Currently, it cannot recognize their parameters and return values but will be able to in a future update.