Agent Chat

Early Alpha Version

AI Agents are in early alpha. Node interfaces, behaviors, and APIs are subject to change.

The Agent Chat node sends messages to an agent and receives responses. It supports both standard request/response and streaming for real-time output.

Inputs

Input	Type	Default	Description
Agent	Object	—	Connect from an AI Agent node
Message	String (multiline)	—	The message to send to the agent
Send	Signal	—	Trigger to send the message
Thread ID	String	—	Optional conversation thread identifier for multi-turn conversations
Streaming	Boolean	false	Enable real-time token streaming
System Override	String (multiline)	—	Override the agent's system prompt for this message only
Cancel	Signal	—	Cancel an in-progress generation

Outputs

Output	Type	Description
Response	String	The complete response text (available when generation finishes)
Partial Response	String	Partial text that updates in real-time during streaming
Finished	Signal	Fires when the response is fully received
Is Generating	Boolean	true while a response is being generated
Token Usage	Object	Token count breakdown: promptTokens, completionTokens, totalTokens
Tool Calls	Array	List of tool calls the agent made during this response
Error	String	Error message if the request failed

Basic Usage

1. Connect an AI Agent to the Agent input
2. Wire a text source to the Message input
3. Wire a button or event to the Send signal
4. Read the result from the Response output

Streaming

When Streaming is enabled, the Partial Response output updates in real-time as tokens arrive. This creates a "typing" effect similar to ChatGPT.

• Connect Partial Response → Text node for live display
• The Response output still contains the full text after generation completes
• The Finished signal fires once all tokens have been received

Thread ID

The Thread ID input enables multi-turn conversations. Messages with the same Thread ID share conversation context.

• Leave empty for single-turn (stateless) interactions
• Set to a unique string (e.g. user ID) for persistent conversations
• Combine with an Agent Memory node for full conversation persistence

System Override

The System Override input lets you temporarily change the agent's behavior for a single message without modifying the Agent node's instructions.

This is useful for:

• Dynamic context injection ("The user is on the checkout page")
• Role switching ("Respond as a translator for this message")
• Adding real-time data to the prompt

Usage Tips

• The Is Generating output is useful for showing a loading spinner or disabling the send button
• Token Usage helps you monitor costs — check totalTokens to track usage
• Tool Calls shows which tools the agent invoked, useful for debugging