Agent Builder

The Agent Builder is a powerful tool that lets you create fully functional agent flows in minutes—without needing to manually configure every step. It uses the same agentic framework as the Agent step, enabling AI agents that can reason, plan, and act based on your instructions. However, Agent Builder automates the process of flow creation, saving you time and effort.

Instead of manually defining system prompts, functions, and handler steps, you simply describe what the agent should do. Agent Builder translates your instructions into a fully configured agent flow, complete with the appropriate functions, data capture, and validation logic.

Note: Agent Builder is ideal for quickly generating production-ready agent flows and is recommended over manual setup using Agent steps when possible.

Before You Start

Before building an agent, make sure you have clarity on:

Agent persona – What is the role of your agent? What task should it complete?
Required tools – What functions should the agent use (e.g., saving an appointment, validating inputs)?
User input – What information should the agent gather from the user?
Validation logic – What checks should the agent perform on the data collected?

Create an agent flow

To create an agent using Agent Builder

Go to your solution flows.
Click the Create new agent button in the top-right corner of the Flows page. The Agent Builder chat opens.
Provide instructions for the agent, either via text or conversational. Describe its goal and any specific behavior.
Specify what tools the agent should use and what each tool is responsible for.
Define the user inputs the agent should collect and indicate what validations are needed (or not).
Review your inputs, and then confirm to create the agent.

The Agent Builder will automatically generate an agent flow based on your instructions. This includes:

A configured Agent AI step with the system prompt and tools. The Run agent integration automatically selected on the Agent AI step is one of the Large Language Model (LLM) integrations defined on entity [[GPT]]. The model used in this integration should support tooling and function calling to ensure the agent can effectively interpret your instructions and utilize the functions you've provided.
A Run Agent step to execute the agent.
Conversational and function call branches, including all necessary function handlers.

The new agent is now ready to chat.