Create an Agent
Every agent in Act starts from a template. A template is a data-defined agent (DDA): a single configuration that describes everything the agent senses, reasons about, acts on, and reflects on. You do not write code. You fill in a template and deploy it.
You can get a starting template in two ways:
-
Clone a Marketplace template: Select a ready-made agent that HiveMQ built, then adapt it. This path is the fastest way to a working agent. See Start from a Marketplace Template.
-
Draft a template with Bea: Open the Create Agent dialog from the Act header and describe your goal in plain language. For example, "Draft an agent to monitor temperature and alert on anomalies". Bea then generates a first-draft template for you to refine.
Either path opens the Agent Builder, the editor where you refine the template before you deploy it.
The Agent Builder
The Agent Builder shows Stages and a form panel. The Stages panel highlights the active stage and shows a dot beside it. A checkmark marks each completed stage. As you make changes, an Unsaved changes badge appears next to the agent name.
|
You can also open the Ask Bea panel to refine any stage in plain language as you work. Close the AI assistant panel to give the form more room. |
From the builder, you can save the agent template at any time with Save template. When all required fields are valid, you can deploy the agent with Deploy agent. After you save the template, you can click the menu icon (the three dots) to download the template as YAML.
Stages in the Agent Builder
The builder has seven stages. Each stage maps to a goal you work toward, and most stages have their own page in this guide.
| Builder Stage | Your Goal | Where to Configure It |
|---|---|---|
01: Agent Configuration |
Give the agent its identity, parameters, and AI model |
|
02: Connections |
Connect the agent to the brokers, databases, and services that it uses |
|
03: Sense |
Collect the live data that the agent observes |
|
04: Reason |
Decide what the data means and what actions to take |
|
05: Actuate |
Carry out the chosen actions in the real world |
|
06: Reflect |
Save memory and metrics to improve the next cycle |
|
07: Termination |
Decide when the agent stops running |
The Sense, Reason, Actuate, and Reflect stages form the four-stage loop of the agent. The agent senses live data, reasons about the data, actuates a response, and reflects on the outcome before the next cycle begins.
Set When the Agent Runs
Each cycle of the loop starts when a trigger fires: on an interval, on a schedule, or when a message arrives. Configure the trigger in the Agent Configuration stage. See Set a Trigger to Run an Agent Cycle.
Add Human Oversight
You decide how much independence the agent has. The range spans full autonomy to human approval for every action. Configure the oversight in the Agent Configuration stage. See Set Human Oversight.