Create an Agent

You can create a custom Agent from a blank configuration using the Agent Builder. From the builder, you can edit inputs and outputs, add nodes as steps, and run tests.

Click “Agents” in the left navigation bar, then click “+ New Agent” to open an empty Agent in the editor.

Configure Inputs

All agents begin with an Input block. Agents use inputs as initial variables that are passed through the first steps of the Agent. Input variables remain accessible to all later steps.

Add Inputs

Click the Input step to configure inputs. Clicking any step opens the step menu. Select “+ Add input” to add new inputs to the Agent.

When adding an input, you must configure three settings:

• Label - The name you will use to reference the input variable in later steps.

• Type - Inputs can be either number or text.

• Required - Specifies whether the Agent requires this input or whether it is optional.

Add Nodes as Steps

After defining inputs, you can begin adding new nodes as steps. Each step represents an action and must be placed in the correct order within the Agent.

You can add a node by clicking the “+” icon in the Agent Editor or by dragging an Agent action from the left navigation bar into the editor.

Each node has its own required inputs and outputs. For example:

• Prompting an answer engine requires you to configure the prompt and label the response as an output.

• Retrieving data from Answer Engine Insights requires specifying the report type, metrics, and other relevant parameters.

Use Variables in Steps

Variables allow you to pass information between steps. You can reference variables as part of a step’s inputs, and you can generate new variables from a step’s outputs.

Use a Variable as an Input

Any input field that accepts variables includes an “(x) Add variable” icon. Click this icon and select the desired variable to include it in the field.

You can also combine static text with a variable. For example, when creating a prompt for an answer engine, you might use an input such as:

“Please write a short description of {custom variable}.”

Label Output Variables

You must label output variables so they can be referenced in later steps. All steps that produce outputs include an Output Label field in their configuration.

Use clear, descriptive variable labels. These become the variable names you select when configuring subsequent steps. 

Add Conditionals

Conditionals are decision points that branch your Agent workflow based on variable values.

Learn how to add and configure Conditional nodes, define comparison expressions, and set up multi-condition logic to route your Agent down different paths depending on inputs or previously generated values.

Add Iterations

Iterations let your Agent repeat the same sequence of steps across every item in a list — taking multiple inputs and producing multiple outputs in a single run.

Learn how to add and configure Iteration nodes, define your list input, build steps inside the loop, and structure your output for bulk processing use cases.

Final Output

The final step in an Agent is the Output step. Unlike outputs generated by individual steps, the final output signifies the completion of the entire Agent.

The Output step can return a final result, such as generated content. Alternatively, the key action of the Agent may occur in an earlier step, with the Output step simply marking the end of the run.

You can add a final Output step and end the Agent by selecting the "End" node.

Publish a Agent

New Agents start as a draft. This allows you to make edits and indicate that the Agent is in testing. Once the Agent is in a production ready state, simply click “Publish” to make the Agent live.