An AI Agent can use a connected chat model, optional memory, tools, a fallback model, and a structured output parser. These are attached to the agent on the workflow canvas, while the agent’s instructions and behavior are configured in its own panel.
Screenshot to add: AI Agent Parameters panel showing prompt source, output-format option, and available attachments
Build the agent
- Add an AI Agent node to the workflow.
- Connect a Chat Model. This is required for the agent to run.
- Open the agent’s Parameters tab and choose the Source for Prompt (User Message). Select the directly connected Chat Trigger source to use its
chatInputvalue, or choose Define below to write a prompt in the node. - Write clear instructions. Add the optional System Prompt from Add Option if the agent needs persistent guidance.
- Add a Tool attachment if the agent needs to call configured actions. Add Memory if the agent needs conversation context.
- If required for your task, enable Enable Fallback Model and connect an additional chat model.
- If the response must follow a JSON structure, enable Require Specific Output Format and connect a Structured Output Parser.
- Test the agent and review its response and node results before using it in a live workflow.
Screenshot to add: AI Agent canvas with connected Chat Model, Tool, Memory, Fallback Chat Model, and Structured Output Parser attachments
Configure execution behavior
Open the agent’s Settings tab to review execution options such as Always Output Data, Execute Once, Retry On Fail, and On Error. When retries are enabled, the panel also exposes retry count and wait settings. Change these options only when they match the workflow’s intended error handling.
Validate the graph before running
The editor validates required attachments and configuration. If the agent cannot run, check the validation message, confirm that the chat model is connected, and resolve any errors on the agent or its attached nodes. Optional attachments should be added only when the workflow needs them.