Skip to main content

Collect Node

Overview​

The Collect Node is utilized to interactively collect data from users via configured slots, which capture necessary information within the workflow.

Configuration​

Adding and Editing Slots​

Users can define slots to collect specific data from user inputs. Each slot can be customized with the following attributes:

  • Label: A user-friendly name. (Also fallback for Description)
  • Key: A unique identifier used programmatically to reference the collected data.
  • Type (optional): Specifies the data type of the slot. Choices include:
    • String (default)
    • Number
    • Enum for predefined values
  • Pattern (Regex) (conditional): A regular expression for validating data, applicable if the slot type is not Enum.
  • Options (Enums) (conditional): Specifies allowed values for the slot, required if the type is Enum.
  • Description (optional): Provides a brief instruction for the LLM on how to handle the slot, aiding in generating appropriate user prompts.

Settings​

Configurable settings include:

  • Connection: Selects the model connector for processing the data
  • Model (LLM): Specifies the large language model to use for processing the data. Model availability depends on the selected connection.
  • Assistant ID (conditional): Required if using a specific assistant configuration with models like gpt4-assistant.
  • Language (optional): Sets the default language for interactions. Not available in Custom Prompts
  • Custom Prompt (optional)

Custom Prompt​

Provides a JavaScript function that customizes the System Prompt handed to the configured Model. It uses the following parameters:

  • items[]: An array of the configured slots within the node.
  • globalData: Contains all previously collected data from other nodes. (object)
  • tone: Can be impersonal, formal, or personal, directing the tone of the interaction.

Example of a custom prompt function:

(({items, globalData, tone}) => `Always format your response in JSON with the schema {slots: {[id]: value}, message, status}. 
    Request the following slots from the user: 
    ${JSON.stringify(items)} while keeping a ${tone ?? "formal"} tone. 
    If done, status is DONE.`)({items, globalData, tone})
note

It is generally advised to use the proven default prompt.

Output Schema​

The custom JavaScript function should output an object adhering to this schema:

  • slots: Key-value pairs where keys are slot IDs, and values are the data collected for those slots.
  • message: Output to user.
  • status: Indicates whether the interaction is complete (DONE) or ongoing (IN_PROGRESS).

Example responses​

The following displays valid responses, resulting from input handling.

Request name​
{
  "slots": {},
  "message": "Please tell me you name.",
  "status": "IN_PROGRESS"
}
Confirm input​
{
  "slots": {"name": "Jon Doe"},
  "message": "Thank you!",
  "status": "DONE"
}

Output​

Emits the results from data collection to subsequent nodes.

Language​

Language set within conversations will always be output to globalData slot _language