> For the complete documentation index, see [llms.txt](https://docs.mydbsync.com/cloud-workflow/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.mydbsync.com/cloud-workflow/ai/ai-agent.md).
# AI Agent
### Overview
The **AI Agent** action lets you apply Large Language Model (LLM) driven actions on incoming data streams. You can provide instructions or prompts, optionally use variables from the **fx** panel, preview the output, refine prompts, and save the final instruction set to execute at runtime.
### Configure and run the AI Agent action
To configure and run the **AI Agent** action, follow these steps:
{% stepper %}
{% step %}
On the **Flow** tab, drag the **AI Agent** action from the action palette and drop it onto the **Flow** panel so that it becomes a sub-action of the flow.
{% endstep %}
{% step %}
Click **Configure**.
{% endstep %}
{% step %}
Optional: Enter a custom name for this step, such as `Get SF lead summary`.
{% endstep %}
{% step %}
In the **LLM** list, select a connected LLM app or connector.
{% endstep %}
{% step %}
In the **Model** list, select an available model.
{% endstep %}
{% step %}
In the **Batch Size** list, select a value: **1**, **10**, **50**, **100**, **200**, or **500.**
{% endstep %}
{% step %}
Click **Configure** to open the prompt setup and preview window.
{% endstep %}
{% step %}
If a prompt isn't configured, this default screen appears. Enter a prompt and click **Submit**  to process it and view the preview.
{% endstep %}
{% step %}
In the **User Prompt** section, you can enter:
* Plain text instructions
* Multi-line prompts
* Data variables inserted via **fx** panel (for example, `{{name}}` or `{{date}}`)
{% endstep %}
{% step %}
Optional: To configure models and tools to your agent, select from the following available options:
* In the **LLM Model** list, you can view the LLM app that you selected on the previous page. You can also select a different LLM connector from the list.
* **MCP Server:** Select an MCP server from the list, or click **Add New MCP Server** to connect a new one.
* **Vector Database:** Select a connected vector database, and then select a vector store or table from the **Collection** list.
* **Web Search:** Configure parameters to enable web search. This feature lets you scope and localize web searches to improve relevance, compliance, and intent-matching by constraining search domains and applying geographic and contextual filters.
{% endstep %}
{% step %}
Optional: To insert variables into your prompt, expand the **fx** panel to view workflow variables and the structure of the incoming data stream. Drag a variable and drop it into the instruction box. Variables appear in curly braces (for example, `{{name}}`).
{% endstep %}
{% step %}
Click **Preview** to process the prompt and generate a preview. The **Preview Output** panel displays the generated output and refreshes automatically whenever you change the instructions.
{% endstep %}
{% step %}
Click **Save & Close**.
{% endstep %}
{% step %}
Click **Run** 
to initiate the integration. The console appears at the bottom of the screen, displaying logs and operations. When the process completes successfully, the console status updates to **Process Completed** and displays the **Data In** and **Data Out** details.
{% endstep %}
{% endstepper %}