Skip to main content

​
What does it do?

Activities nodes define the core processing steps that your assistant performs. They contain the instructions and configuration that determine how your assistant processes inputs and generates outputs. Think of Activities as the “brain” of your workflow - where the actual work happens. Activities take information from Input nodes and transform it according to your instructions and requirements.

​
How do I use it?

​
Setting Up an Activity Node

When you add an Activity node, you’ll first choose from pre-configured activity types, then configure how your assistant should perform that specific activity:

​
Pre-Configured Activity Types

Answer Questions: Responds to user inquiries by analyzing the question and providing relevant answers
  • Prompt purpose: Guide HOW to answer questions (tone, depth, sources to reference, format)
  • Example prompt: “Answer customer questions in a friendly, professional tone. Always check the knowledge base first and escalate complex technical issues to @tech support.”
Summarize Content: Condenses information into concise summaries while preserving key points
  • Prompt purpose: Define HOW to summarize (length, focus areas, format, what to emphasize)
  • Example prompt: “Create 3-4 bullet point summaries focusing on key business insights and actionable recommendations. Include relevant metrics when available.”
Sentiment Analysis: Analyzes the emotional tone and sentiment in text content
  • Prompt purpose: Define HOW to analyze sentiment and present findings
  • Example prompt: “Analyze customer feedback for sentiment (positive, negative, neutral). Highlight key themes and provide actionable insights for improvement.”
Write Emails: Creates email content based on context and requirements
  • Prompt purpose: Specify tone, format, and style for email composition
  • Example prompt: “Write professional customer service emails that are empathetic and solution-focused. Include clear next steps and appropriate sign-off.”
Custom: Flexible activity type for specialized tasks unique to your needs
  • Prompt purpose: Define any custom processing task and how it should be performed
  • Example prompt: “Analyze product reviews and extract key feature mentions, rating trends, and improvement suggestions formatted as actionable insights.”

​
Activity Configuration

After selecting your activity type, configure these components: Custom Prompt: Instructions that guide HOW your assistant should perform the chosen activity type. Response Type: Choose between text responses (natural language) or structured responses (JSON format). Model Selection: Select which AI model should process this activity.

​
Custom Prompt Configuration

Write clear, specific instructions that guide HOW your assistant should perform the selected activity type. Remember: the activity type defines WHAT to do, your prompt defines HOW to do it. Example prompts:
  • “Analyze the customer support ticket and categorize it as: technical, billing, general inquiry, or complaint. Provide a brief summary.”
  • “Extract the customer’s name, email, order number, and issue description from this support message”
  • “Generate a professional email response addressing the customer’s concerns with an empathetic tone”
  • “Summarize the key points from this document in 3-4 bullet points”
Good prompt structure:
  • Start with action words: “Analyze”, “Extract”, “Generate”, “Summarize”
  • Be specific about what you want
  • Include format requirements when needed
  • Mention any constraints or guidelines

​
Response Type Selection

Text Response:
  • Natural language output
  • Good for explanations, customer responses, reports, summaries
  • Easy to read and understand
Structured Response (JSON):
  • Organized data format with defined fields
  • Good for data extraction, categorization, form filling
  • Requires you to define field names and descriptions
When you select “Structured”, you’ll add JSON fields:
  • Field Name: The key in your output (e.g., “priority”, “category”, “summary”)
  • Description: What this field should contain

​
Best Practices

​
Write Clear Instructions

Be specific about what you want the assistant to do. Include examples when helpful. Good: “Extract the customer’s name, email, and order number from this support ticket” Poor: “Process this customer message”

​
Choose the Right Response Type

  • Use Text for human-readable responses like emails, summaries, or explanations
  • Use Structured when you need consistent data format for categorization or extraction

​
Keep Activities Focused

Each Activity should handle one specific task. Break complex processes into multiple connected activities for better results.

​
Test with Real Data

Use actual examples of the inputs you expect to receive to test and refine your prompts.
⌘I