This article introduces a method for implementing a multi-workflow intelligent agent based on openJiuwen using a low-code development approach.

1. Building Workflows

Before constructing the intelligent agent, it is necessary to prepare the corresponding workflows. This article demonstrates the process using two example workflows: Balance Inquiry Workflow and Transfer Workflow.

1. Balance Inquiry Workflow Setup

This workflow is used to simulate a balance inquiry scenario.

• Step 1: On the homepage, click “Create Workflow” to enter the workflow development interface.
  

• Step 2: Enter the workflow name query_balance and the description
  Balance inquiry workflow, which can query account balances under a user’s name,
  then click “Confirm Creation”.
  

• Step 3: Add a Questioner node on the canvas and configure it according to the illustrations.
  
  

• Step 4: In the End node, concatenate the output of the Questioner into the predefined output and enable streaming output.
  

  At this point, the balance inquiry workflow setup is complete.

2. Transfer Workflow Setup

This workflow is used to simulate an account transfer scenario.

• Step 1: Following the same process as the balance inquiry workflow, create a new workflow and enter the canvas.
  

• Step 2: Add a Questioner node and configure the parameters to be asked in the node settings.
  

• Step 3: Connect all nodes properly and enable streaming output in the End node.
  

  At this point, the transfer workflow setup is complete.

2. Building a Multi-Workflow Intelligent Agent

After completing the preparation of the two workflows above, you can start building a multi-workflow intelligent agent.

• Step 1: Enter the intelligent agent development interface and click “Create Agent”.
  

• Step 2: Enter the agent name Financial Customer Service Assistant and the function description
  A financial customer service assistant that can provide complete transfer and balance inquiry services.
  

• Step 3: The default agent mode is Autonomous Planning Mode. Manually switch it to Multi-Workflow Mode.
  

• Step 4: In the orchestration configuration, add the two created workflows, and configure the opening greeting in the conversation settings.

After completing the above steps, a financial customer service assistant based on multi-workflow mode is successfully built.

3. Effect Testing

Now let’s test the results.

First, enter the question I need to make a transfer in the chat box on the right. The agent correctly uses the Questioner in the transfer workflow to ask the user relevant questions.

Next, input Let me check my balance first. You can see that the agent exits the current workflow and switches to the balance inquiry workflow, asking the user for the name to be queried.

Then, tell the agent My name is Zhang San, check my balance. The agent successfully resumes the current workflow and completes the balance inquiry.

Finally, input Transfer 100 yuan to Li Si. The agent restores the previous workflow state and completes the transfer operation.