AI Assistant in Pipeline Nodes

​

What is AI Assistant?

• Generate content from natural language instructions
• Improve existing content by providing refinement prompts
• Iterate quickly with split-view comparison between original and improved versions
• Leverage AI reasoning to create better configurations

​

How AI Assistant Works

1. Editor Area - View and edit your content with syntax highlighting
2. AI Prompt Input - Natural language instructions for content generation
3. Language Selector - Choose syntax highlighting (auto-detected)
4. Action Buttons - Copy, apply, and close controls

• Editor displays your current content (or empty state)
• Prompt Input accepts your instructions
• Generate creates content based on your prompt
• Result appears directly in the editor

• Left Panel - “Current Version” (read-only reference)
• Right Panel - “Improved Version” (AI-generated result)
• Compare - See differences side-by-side
• Apply - Replace current content with improved version
• Close - Discard improvements and return to single view

​

Supported Nodes and Fields

​

LLM Node

• Generate comprehensive system prompts for specialized agents
• Create clear, structured task instructions
• Build example conversations for testing

You are an expert software testing assistant specializing in unit test creation. Your role is to:

1. Analyze code snippets provided by developers
2. Identify testable units and edge cases
3. Generate comprehensive unit tests with clear assertions
4. Follow best practices for the testing framework in use
5. Explain your testing strategy and coverage decisions

Provide tests that are:
- Clear and maintainable
- Cover happy paths and edge cases
- Include meaningful test names and descriptions
- Follow the AAA pattern (Arrange, Act, Assert)

​

Code Node

• Generate data transformation logic
• Create calculations and business rules
• Build API integration code
• Improve code quality and error handling

def calculate_discount(order_value):
    """
    Calculate tiered discount based on order value.
    
    Args:
        order_value (float): Total order value
        
    Returns:
        dict: Discount percentage and final price
    """
    # Get order value from state
    order_value = float(alita_state.get('order_value', 0))
    
    # Determine discount tier
    if order_value >= 1000:
        discount_rate = 0.20
    elif order_value >= 500:
        discount_rate = 0.15
    elif order_value >= 100:
        discount_rate = 0.10
    else:
        discount_rate = 0.0
    
    # Calculate discount and final price
    discount_amount = order_value * discount_rate
    final_price = order_value - discount_amount
    
    return {
        "discount_rate": discount_rate,
        "discount_amount": round(discount_amount, 2),
        "final_price": round(final_price, 2)
    }

# Execute and return results
result = calculate_discount(order_value)
return result

​

State Modifier Node

• Format output messages with state variables
• Build reports and summaries
• Build structured responses

# User Story: {{ title }}

## Description
{{ description }}

## Acceptance Criteria
{% for criterion in acceptance_criteria %}
- {{ criterion }}
{% endfor %}

## Metadata
- **Priority**: {{ priority }}
- **Story Points**: {{ story_points }}
- **Sprint**: {{ sprint }}
- **Assignee**: {{ assignee }}

---
*Created on {{ created_date }} by {{ author }}*

​

Router Node

• Create multi-path routing logic
• Build approval workflows
• Implement status-based routing

{% if 'approved' in status|lower %}
Approved
{% elif 'rejected' in status|lower %}
Rejected
{% elif 'review' in status|lower or 'pending' in status|lower %}
Pending
{% else %}
Default
{% endif %}

​

Decision Node

• Describe decision criteria clearly
• Provide context for intelligent routing
• Document decision-making logic

Analyze the customer support ticket and route it to the appropriate team based on these criteria:

**High Priority (Critical Issues):**
- System outages or downtime
- Security vulnerabilities
- Payment processing failures
- Data loss or corruption

Route to: **CriticalSupport**

**Billing & Payments:**
- Invoice questions
- Payment issues
- Subscription changes
- Refund requests

Route to: **BillingTeam**

**Technical Support:**
- Feature questions
- Integration issues
- API problems
- Bug reports

Route to: **TechnicalSupport**

**General Inquiries:**
- Account questions
- Feature requests
- General questions

Route to: **GeneralSupport**

**Default:** Route to **GeneralSupport** if unsure

​

HITL Node

• Craft clear, context-aware approval request messages
• Build F-String messages that embed state variable content for user review
• Write actionable interrupt prompts that guide the user on what to do next
• Create professional review messages for compliance or content approval workflows

Please review the article draft before it is published.

**Title:** {article_title}

**Summary:**
{article_summary}

Choose an action:
- **Approve** — publish the article as-is
- **Edit** — modify the draft content before publishing
- **Reject** — discard the draft and restart generation

​

Printer Node

• Display static completion messages
• Show formatted output combining text with variables
• Pause pipeline for user inspection
• Show progress updates during multi-step workflows

==========================================
Processing Complete: {project_name}
==========================================

Results:
  • Total Items: {total_items}
  • Successful: {success_count} ({success_rate}%)
  • Failed: {failed_count}
  • Processing Time: {processing_time}s

Status: {status_indicator}

{completion_message}
==========================================

==========================================
Processing Complete: Customer Analytics Dashboard
==========================================

Results:
  • Total Items: 150
  • Successful: 142 (94.7%)
  • Failed: 8
  • Processing Time: 12.5s

Status: ✓

Processing completed successfully with high success rate.
==========================================

​

Using AI Assistant: Step-by-Step

​

1. Generate New Content

1. Open the LLM node configuration
2. Click the AI Assistant icon (🪄) on the “Task” field
3. Enter your instruction in the AI Prompt Input:
   Create a system prompt for a code reviewer that focuses on security best practices
   
4. Click “Generate” or press Enter
5. Review the generated content in the editor and edit if needed
6. Close the modal - content is now in the System field

​

2. Improve Existing Content

1. Open AI Assistant on a field with existing content
2. Review current version in the editor
3. Enter improvement instruction in the AI Prompt Input:
   Make this prompt more specific for reviewing API security in REST endpoints
   
4. Click “Generate” - AI Assistant switches to split view
5. Compare versions side-by-side:
   • Left: Original content
   • Right: Improved content
6. Click “Apply” to use the improved version, or
7. Click “Close” (X) to discard and keep original

​

3. Iterative Refinement

1. Generate initial content using AI Assistant
2. Review the result in split view
3. Enter a refinement prompt without applying:
   Add examples of common security vulnerabilities to look for
   
4. Click “Generate” again - improved content updates
5. Continue refining with additional prompts:
   Make the tone more conversational and encouraging
   
6. Apply when satisfied with the final result

​

4. Error Handling and Retry

• Error message displayed in the editor
• Prompt input remains filled with your instruction
• Split view (if previously shown) stays open

1. Review the error message (e.g., “Error: Rate limit exceeded”)
2. Your prompt is preserved - no need to retype
3. Click “Generate” again to retry
4. AI Assistant retries with the same prompt
5. Previous error is cleared if successful

​

Common Use Cases

Prompt: "Create a system prompt for an AI assistant that generates SQL queries from natural language. It should only output valid SQL, never execute queries, and explain any assumptions."

Result: Detailed system prompt with role definition, constraints, output format, and safety guidelines.

Prompt: "Create router condition for approvals: route to 'AutoApprove' if amount < 1000, 'ManagerApproval' if 1000-10000, 'ExecutiveApproval' if >10000, default to 'ReviewQueue'"

Result: Properly structured Jinja2 condition with numeric comparisons and fallback.

Prompt: "Write Python code that takes a list of user dictionaries and creates a summary with total users, breakdown by role, and list of admin emails"

Result: Working Python code with error handling, state access, and structured output.

Prompt: "Create a Jinja template for a daily report showing completed tasks, pending tasks, overdue tasks, and team member performance"

Result: Professional template with formatting, conditionals, and loops.

Prompt: "Write an F-String message for a compliance review checkpoint that shows the generated policy text and explains what approve, edit, and reject will each do"

Result: Structured review message with {state_variable} placeholders, plain-language button descriptions, and a professional tone suitable for compliance workflows.

​

Best Practices

​

Troubleshooting