Automating Workflows

Automating workflows in AgentScope AI streamlines complex processes by enabling agents to execute predefined sequences of actions with minimal human intervention. Workflows integrate multiple tools, data sources, and decision-making logic to efficiently complete tasks. This section provides a detailed guide on defining, managing, and executing automated workflows within AgentScope AI.

Creating an Automated Workflow

To create an automated workflow, you need to:

1. Define the workflow steps.

2. Specify the tools and agents involved.

3. Configure triggers and conditions.

4. Implement execution logic.

Example: Automating a Support Ticket Workflow

In this example, we create an automated workflow where an AI agent categorizes incoming support tickets and assigns them to the appropriate team.

Step 1: Define the Workflow Schema

import { createWorkflow } from "@agentscope/core";
import { z } from "zod";

export const supportTicketWorkflow = createWorkflow({
  id: "Support Ticket Handling",
  inputSchema: z.object({
    ticketId: z.string(),
    description: z.string(),
  }),
  description: "Automatically categorizes and assigns support tickets based on content analysis.",
});

Step 2: Integrate AI Agent for Categorization

import { Agent } from "@agentscope/core";
import * as tools from "../tools/textAnalysis";

export const supportAgent = new Agent<typeof tools>({
  name: "Support Categorization Agent",
  instructions:
    "Analyze support tickets and determine the appropriate category based on the issue description.",
  model: {
    provider: "OPEN_AI",
    name: "gpt-4",
    toolChoice: "required",
  },
  tools: {
    textAnalysis: tools.textAnalysis,
  },
});

Step 3: Define Workflow Execution Logic

supportTicketWorkflow.execute = async ({ context }) => {
  const { ticketId, description } = context;
  console.log(`Processing ticket ${ticketId}`);

  const category = await supportAgent.generate(
    `Categorize this support ticket: ${description}`
  );
  
  return { ticketId, category };
};

Managing Workflow Execution

Once defined, workflows can be executed either manually or triggered automatically based on external events.

Manual Execution

import { runWorkflow } from "@agentscope/core";

async function main() {
  const result = await runWorkflow(supportTicketWorkflow, {
    ticketId: "12345",
    description: "My account is locked. How can I reset my password?",
  });
  console.log(result);
}

main();

Automating Execution via Webhooks

import { createWebhookTrigger } from "@agentscope/core";

export const webhookTrigger = createWebhookTrigger({
  workflow: supportTicketWorkflow,
  endpoint: "/api/support-ticket",
});

Monitoring and Debugging

AgentScope AI provides logging and debugging features to monitor workflows:

• Execution Logs: Track each step's inputs and outputs.

• Error Handling: Define fallback mechanisms for failed executions.

• Performance Metrics: Analyze execution times and optimize workflows accordingly.