Agent

Important: Agents is currently in private beta and is not yet available to all customers.

Use the Agent action to configure and run an agent within an agentflow. The action allows you to define the model, connection, parameters, role, and task instructions. The agent generates a response based on the configuration and stores the result in an output variable.

You can use this action to automate content generation, structured reasoning, classification, or task execution within an agentflow.

For example, a support team receives incoming requests and wants to classify them and draft a response. Configure the Agent action with defined role instructions and a work request. The agent processes the input and returns a response that can be stored in a variable and used in subsequent steps of the agentflow.

Configure the Agent action

1. Add the action to the agentflow and open the action configuration panel. For more information, see Add an action to the agentflow.

2. The configuration is organized into four tabs: Model, Tools, Knowledge, and Output

Configure the Model

Use the Model tab to define the model and behavior settings.

1. Select a model from the list in the Model field.

2. In Model Parameters, configure the following:

   1. Connection: Add or select a connection used to access the model.

   2. Model Name: Specify the model version to use (for example, gpt-4).

   3. Reasoning effort: Select the reasoning level you want the agent to apply. Choose Low, Medium, or High from the list. This option is available only for o1 and o3 models.

   4. Temperature : Specify the temperature value to control response variability. Lower values produce more predictable results. Higher values produce more varied responses.

3. In Agent role and rules, define the agent’s role, behavior, and constraints.

4. In Work request, describe the task or question the agent should respond to.

For more information about the model and its parameters, see Model and parameters.

Configure Tools

Use the Tools tab to select tools that extend the agent’s capabilities.

1. Open the Tools tab.

2. In Choose tools, select one or more tools the agent can use.

   For example, select Calculator if the agent needs to perform mathematical calculations.

3. In Configure tools, configure the selected tools as required.
   Select tools that are only necessary for the agent’s task.

For more information about tools available in the agent, see Tools

Configure Knowledge

Use the Knowledge tab to define what information the agent can reference.

1. Open the Knowledge tab.

2. In Shared context, select whether the agent can access outputs from earlier agents in the agentflow.

   Note: Enabling shared context increases token usage.

3. In Document store, select or add a document store.

4. In Vector embedding, select or add vector embeddings that contain structured knowledge for the agent.

   Use knowledge sources when the agent must respond based on specific documents or stored data.

For more information about document store and vector embedding, see Knowledge.

Configure the Output

Use the Output tab to define how the response is returned.

1. Open the Output tab.

2. In Return response as, select how the response should be structured.

   For example, select User message to return the response as standard text.

3. In Output, select or create a variable to store the agent’s response.

   The stored output can be used in later steps of the agentflow.

For more information on output, see Output.

Agent action fields and settings

Section	Field	Description	Variable
Model	Model	The model provider used by the agent. For example, Azure ChatOpenAI, ChatAnthropic. For more information, see Model and parameters	N/A
Model Parameters	Connection	The connection used to access the selected model. To refresh the available connections, click .	N/A
	Model name	The model version used to generate the response.	N/A
	Reasoning effort	The reasoning level applied before generating a response. Available only for o1 and o3 models. Select Low for faster, simpler responses, Medium for a balance of quality and speed, or High for more thorough responses.	N/A
	Temperature	The level of creativity in the response. Lower values return more consistent responses. Higher values return more varied responses.	Decimal, Integer,
	Agent role and rules	Instructions that define the agent’s role, expected behavior, and any constraints.	Text, Decimal, Integer, Boolean, DateTime, Collection
	Work request	Instructions describing the task or question the agent should respond to.	Text, Decimal, Integer, Boolean, DateTime, Collection
Tools	Choose tools	The tools available for the agent to perform specific tasks. For more information, see Tools	N/A
	Configure tools	Settings for the selected tools. For more information, see Tools	N/A
Knowledge	Shared context	Allows the agent to reference outputs from earlier agents in the agentflow. Enabling this option increases token usage.	Boolean
	Document store	Document stores that provide reference content for the agent.	Collection
	Vector embedding	Vector embeddings that provide structured reference content.	Collection
Output	Return response as	Defines how the response is structured.	N/A
	Output	Stores the generated response in a variable.	Object