Enter the Application Details page, click Workflow at the top of the page to switch to the workflow page. This page supports management features such as create, configure, enable, delete, import and export workflows.
Step 1: Create Workflow
The Intelligent Agent development platform supports two ways to create workflows: "manual entry" and "batch import". "Manual entry" supports configuring workflows from scratch, while "batch import" supports importing compressed package files to replicate workflows across applications and share workflows among teams.
Manually enter a workflow
1.1 Click Create > Manual Entry, fill in the workflow name and workflow description.
To help the Large Model better understand the workflow's purpose, we recommend using a fixed format for the description. Example:
Description: Help users query the shipping cost of an item(Please describe the specific features of this workflow in natural language)
Example: How much does it cost to mail an ID card to Shenzhen? Is it expensive to send eggs from Beijing to Shenzhen?(Please list common user conversation content that can trigger this workflow)
1.2 After filling in the workflow name and description, click Confirm to complete the workflow creation and enter the workflow canvas page.
Batch import workflows
1.1 Click Batch Import, upload a zip package exported from the Cloud Intelligence Agent Development Platform workflow, with a single file not to exceed 5MB.
1.2 Click Confirm to complete the workflow import and enter the workflow canvas page.
Step 2: Configure Workflow
After entering the workflow canvas page, you can drag and drop workflow nodes to the canvas and connect nodes based on business processes.
1. On the left side of the page, select the necessary nodes from the node panel and add them to the canvas via mouse drag.
2. For two connected nodes, if you wish to insert another node between them, hover the mouse over the connection, click the "➕" icon and select the node to insert.
3. Connect nodes on the canvas according to your business requirements. The connection method between nodes can be divided into serial and parallel connection. The input variable of a node can refer to the output variables of ALL ancestor nodes.
Example of serial connection: The input and output of each node is connected by only one connecting line. As shown in the figure below, the four nodes are exactly serially connected.
Example of parallel connection: The input or output of a node is connected to multiple connecting lines.
4. Click a node on the canvas, and the detailed configuration page will slide out on the right side of the page, where you can configure the input variable, handling process, and output variable of that node.
Step 3: Debug Workflow
During workflow configuration, you can debug a single node at any time. Once configured, you can perform end-to-end testing. The Intelligent Agent development platform workflow supports debugging through multi-round dialogue and shows the detailed execution result of each node.
Single node debugging
1.1 For partial nodes, you can debug node functions through the single node debugging feature. Nodes with this capability will display a "Debug" icon in the top-right corner of the details page (as shown below).
1.2 Fill in the required input parameters of the node in the debug window, then click Debug.
1.3 After the trial run, the execution result will be shown on the page.
End-to-end debugging
1.1 After completing the configuration, click Debug at the top right of the page to enter the end-to-end workflow debugging page.
Note:
If your workflow references API parameters, you can simulate these API parameters through the "set parameters" feature in the test window. During workflow debugging, the information will be entered into the workflow.
1.2 Enter your test content in the input box on the debugging page, then press the Enter key or click Send to send the information. The system will run the workflow and output results according to the content you sent.
1.3 During debugging, the canvas shows the status of the current node (including debugging, success, failure), input/output variable details for each node operation, and the status of connecting lines. You can click the content of the system reply in the debug window to view the path of its generation in the canvas.
Step 4: Use Workflow in Application
1. After configuring the workflow, return to the flow management page to enable or disable it.
Enable: means the workflow can be normally called. You can enable it after debugging.
Disable: means the workflow cannot be normally called. You can disable it when you want to continue configuration or optimize the workflow (status is "disabled").
2. After enabling the workflow (status is "awaiting release" and enabled status is "enabled"), you can test the application in the test window. The application can normally call the workflow, and detailed information of the workflow call can be seen during the test.
