Agent
The Agent node sends a message or instruction to one of your configured AI agents and returns its response. Use it to delegate tasks to agents that have access to tools, knowledge bases, and other capabilities.
Configuration
| Field | Required | Description |
|---|---|---|
| Agent | Yes | The AI agent to use. Select from the dropdown list of agents configured in the project. |
| Message | Yes | The instruction or message to send to the agent. Type @ to insert variables from previous steps. |
| Response Format | No | The format of the agent's response. Default is Normal. |
Chat Settings
| Field | Required | Description |
|---|---|---|
| Create New Chat | No | When enabled, creates a new chat conversation for this agent interaction. Use it when you want each workflow run to start a fresh conversation with the agent. |
| Ignore Chat History | No | When enabled, the agent ignores previous chat history for this interaction. Use it when the agent's response should not be influenced by previous interactions. |
| Chat Name | No | An optional name for the chat conversation. Type @ to insert a variable or enter a name directly. |
| Chat ID | No | An optional custom chat ID. Overrides the auto-generated chat ID when provided. Use it to continue an existing conversation across multiple workflow runs. |
Creating a new chat
When an agent runs a workflow, the result goes back to the agent, which writes the reply in the chat where the workflow was triggered. Create New Chat posts a reply in a new chat at the same time.
Add a Final Output node to choose what the workflow returns. That return value is what the LLM uses to write the reply. Without a Final Output node, the agent receives the full workflow result, including any answer the workflow already posted. A prompt can guide the wording, but it does not choose what is sent back.
Documents
| Field | Required | Description |
|---|---|---|
| Documents | No | Document keys to include in the agent's context. Enter comma-separated keys or type @ to select variables. |
Results
The Results tab shows the agent's response after the node is executed.