AI Flow Assistant
The AI Flow Assistant is an intelligent helper that can create, modify, and troubleshoot your conversational flows using natural language commands.
Important Disclaimer: The AI Flow Assistant generates responses using artificial intelligence. While it includes validation and automatic error correction, responses may contain errors or unexpected results. Always review proposed changes before applying them to your flow.
Accessing the Assistant
Click the AI Assistant button (gradient-colored floating button) in the bottom-right corner of the flow editor to open the chat panel.
What the AI Can Do
Create Nodes
The assistant can create any of the 27+ supported node types:
| Category | Node Types |
|---|---|
| Messages | Message, File, Goal |
| User Input | Ask Question, Ask Email, Ask Phone, Ask Date, Ask File, Ask Location |
| Interactive | Interactive Options, Global Keywords, AI Router |
| Logic | Conditional, Set Variable, Formula, Python Executor |
| Integration | HTTP Request, Workflow Execute |
| Authentication | Login, Logout, OTP Verification |
| Verification | Liveness, OCR, Point of Contact |
| Entity | Create Borrower, Read Borrower, Update Borrower |
| Navigation | Jump To, End, Human Takeover |
Connect Nodes
The AI can create connections between nodes while respecting all connection rules automatically.
Modify Configuration
- Change message content and formatting
- Update questions and associated variables
- Modify interactive options and buttons
- Adjust conditional logic and operators
- Configure HTTP requests (URL, method, headers, body)
- Set up workflow executions
Manage Variables
- Create new variables with appropriate types
- Delete unused variables
- Suggest variable names based on context
Create Complete Flows
The assistant can build entire flows from scratch, including:
- Onboarding flows (individual or company)
- Verification flows
- Data collection flows
- Custom flows based on your description
Troubleshoot Errors
- Analyze validation errors in your flow
- Suggest and apply fixes automatically
- Explain why errors occur
Explain Your Flow
Ask the assistant to:
- Summarize what your flow does
- Explain specific nodes
- Describe the flow logic
What the AI Cannot Do
| Limitation | Reason |
|---|---|
| Connect two Interactive Options directly | Validation rule - requires intermediate node |
| Leave multi-output handles disconnected | Login, Conditional, OCR nodes require all outputs connected |
| Delete the START node | Required entry point for all flows |
| Create invalid node types | Only supports defined node types |
| Use invalid conditional operators | Must use supported operators for each data type |
| Guarantee perfect results | AI-generated content may contain errors |
Technical Limits
- Correction attempts: Maximum 3 automatic retry attempts if validation fails
- Conversation history: Last 20 messages are considered for context
- Processing time: Complex flows may take longer to generate
How to Use
1. Send a Request
Describe what you want in natural language:
Good examples:
- "Add a message node that says 'Welcome to our service'"
- "Create a conditional that checks if user_age is greater than 18"
- "Connect the login node to a message for successful login"
- "Build an onboarding flow for individual customers"
- "Fix the validation errors in my flow"
Less effective:
- "Make it better" (too vague)
- "Add stuff" (unclear what to add)
2. Review Proposed Changes
The assistant shows proposed changes in the Proposals tab:
- Green badge: New element being created
- Blue badge: Existing element being modified
- Red badge: Element being deleted
Each change shows:
- What will change
- Before and after comparison (when applicable)
- Description of the change
3. Apply or Reject Changes
You have full control:
| Action | Description |
|---|---|
| Apply individually | Click ✓ on a specific change |
| Reject individually | Click ✗ on a specific change |
| Apply All | Accept all proposed changes at once |
| Reject All | Discard all proposed changes |
4. Processing Phases
The assistant shows its progress through phases:
- Analyzing - Understanding your request
- Planning - Creating an execution plan
- Reviewing Nodes - Checking node configurations
- Reviewing Edges - Checking connections
- Executing - Applying changes
- Finalizing - Validating results
Best Practices
Be Specific
Instead of: "Add a question"
Try: "Add a question node that asks for the user's email and stores it in the variable user_email"
Review Before Applying
- Always check the "Before" and "After" comparisons
- Verify node connections make sense
- Ensure variables are named appropriately
Use for Complex Tasks
The AI excels at:
- Creating multiple connected nodes at once
- Building repetitive structures
- Fixing multiple validation errors
- Explaining complex flow logic
Iterate if Needed
If the first result isn't perfect:
- Provide feedback on what needs adjustment
- Ask for specific modifications
- The assistant remembers the conversation context
Troubleshooting
Changes not appearing
- Check the Proposals tab for pending changes
- Ensure you clicked "Apply" on the changes
- Refresh the flow if changes seem stuck
Validation errors after applying
- The AI validates changes but some edge cases may slip through
- Use "Fix validation errors" to have the AI attempt corrections
- Manual adjustment may be needed for complex cases
Assistant seems stuck
- Check your internet connection (requires real-time streaming)
- Try refreshing the page
- Cancel the current request and try again
Unexpected node configurations
- Review the change details before applying
- Reject and request with more specific instructions
- Manual editing is always available as fallback