DocsICP FilingConsole
官方文档
首页

Guide to AI-generated flows

更新时间：
复制 MD 格式
产品详情
我的收藏

AI-generated flow is an intelligent flow generation tool powered by the Qwen Large Language Model (LLM). It converts natural language descriptions or structured text into automated multi-turn messages and flow nodes, with built-in preview and editing capabilities. Use AI-generated flows to quickly create automated message flows and accelerate chatbot design for complex business scenarios.

Scenarios

When building automated chatbots for customer service, appointment booking, or food ordering, designing multi-turn conversation flows by manually dragging and configuring nodes on the canvas is tedious and time-consuming, especially for scenarios with complex conditional branches and multiple user interactions.

The AI-generated flow feature addresses this by letting you describe your business needs in plain language. The system uses the Qwen Large Language Model (LLM) to automatically generate a complete, visual flowchart. You can then validate, fine-tune, and deploy the flow in minutes instead of hours. Non-technical users can also create complex flows through simple operations.

Usage Notes

  • This feature is only available in the console. No API is currently provided.

  • After generating a flow, review and adjust it based on your actual business needs to ensure the logic and content meet requirements.

  • This feature has usage limits. Use it wisely.

  • AI-generated results are for reference only. Manually verify content at critical business nodes.

Architecture

The core of the AI-generated flow feature is an intelligent transformation engine that parses unstructured or semi-structured input and builds a structured flow definition for the flow editor.

  1. Input processing: In quick generation mode, enter a natural language description. In professional generation mode, provide structured text in YAML format.

  2. AI parsing and generation: The input is sent to the backend AI generation service. Using the Qwen LLM, this service parses the intent and identifies nodes, logic branches, variables, and interaction steps.

  3. Flow definition construction: The AI service converts the parsed result into a standard flow definition file (such as JSON) that precisely describes each node’s type, content, and connections in the flowchart.

  4. Visualization and editing: The flow definition is returned to the frontend flow editor and rendered as a visual flowchart. You can inspect the logic in the preview interface or load it directly onto the canvas for deep editing and further configuration.

Procedure

  1. Navigate to the feature

    1. Log on to the Chat App Message Service console.

    2. In the navigation pane on the left, click Chat Flow > Create with AI.

  2. Generation flow

    1. Select a mode:

      • Quick Generation: Enter a single-sentence scenario description or select a prompt template to generate a basic flow. Use this mode for simple, clear requirements.

        Example: Help me design a welcome and onboarding flow for new users.

      • Professional Generation: Enter a structured flow description in YAML format or select a prompt template to generate a basic flow. Use this mode for complex scenarios requiring precise control over node behavior and branch logic.

        Example:

        Coupon distribution flow prototype:
1. Send message: "Reply with the number before the event type to get a coupon: 1. BlackFriday; 2. Christmas". Wait for user reply with a timeout of 90 seconds. Based on the reply:
1.1. If the user replies with [1]:
  1.1.1. Send message: "Your BlackFriday coupon code is: *********. Keep it safe!"
  1.1.2. Jump back to step 1. (maximum one jump)
1.2. If the user replies with [2]:
  1.2.1. Send message: "Your Christmas coupon code is: *********. Keep it safe!"
  1.2.2. Jump back to step 1. (maximum one jump)
1.3. If the user does not reply within 90 seconds or sends an invalid reply:
  1.3.1. Send message: "Your reply is invalid or timed out. Please try again!"
  1.3.2. Jump back to step 1. (maximum one jump)
      Note

      • When you select a prompt template, it replaces any existing content in the input box.

      • In Professional Generation mode, after entering a structured flow description, click Prompt Optimization to refine your input.

    2. Click the Generate Preview button and wait for generation to complete. After previewing, you can use these core features:

      • Flowchart visualization: View the entire conversation flow structure clearly.

      • Component details: Click any component to view its detailed configuration parameters.

      • One-click creation: Load the entire previewed flow onto the canvas for manual editing, adding API calls, and other advanced configurations.

  3. Review the flow

    Check whether the AI-generated flow meets your business needs. To make adjustments:

    • In Quick Generation mode, click Refine in Pro Mode in the top-right corner to switch to Professional Generation mode. Modify the content in the Flow Description text box to a structured format, then click Generate Preview.

    • In Professional Generation mode, update the relevant content in the Flow Description text box, then click Generate Preview.

  4. Create the flow

    After verifying the preview, click the Create Flow button to open the flow editor canvas.

  5. Publish the flow

    1. Click the Save button in the top-right corner of the canvas.

    2. Click Settings > Account Binding, select an account, and click OK.

    3. Click the Save and Publish button.

What to do next

  • Use heat map analysis for deeper insights into your message flow performance. Visualize node execution counts, execution percentages, and error rates to optimize your flow design and improve overall response rates. For more information, see Flow Editor Heat Map Analysis and AI-based Intelligent Analysis Guide.

  • Use the data panel to understand how your flow is triggered in practice. Data fields vary slightly by trigger type. For more information, see Flow Editor Data Panel Guide.

References

Refer to the following documents to learn more about flow editor canvas features and component configurations.

Previous：Data panelNext：Intelligent marketing