@aigne/example-workflow-handoff 1.1.0-beta.12

package/.env.local.example

@@ -0,0 +1,5 @@
+# Change the name of this file to .env.local and fill in the following values
+
+DEBUG=aigne:mcp
+
+OPENAI_API_KEY="" # Your OpenAI API key

package/README.md

@@ -0,0 +1,102 @@
+# Workflow Handoff Demo
+
+This is a demonstration of using [AIGNE Framework](https://github.com/AIGNE-io/aigne-framework) to build a handoff workflow.
+
+## Prerequisites
+
+- [Node.js](https://nodejs.org) and npm installed on your machine
+- [OpenAI API key](https://platform.openai.com/api-keys) used to interact with OpenAI API
+- [Pnpm](https://pnpm.io) [Optional] if you want to run the example from source code
+
+## Try without Installation
+
+```bash
+export OPENAI_API_KEY=YOUR_OPENAI_API_KEY # setup your OpenAI API key
+
+npx -y @aigne/example-workflow-handoff # run the example
+```
+
+## Installation
+
+### Clone the Repository
+
+```bash
+git clone https://github.com/AIGNE-io/aigne-framework
+```
+
+### Install Dependencies
+
+```bash
+cd aigne-framework/examples/workflow-handoff
+
+pnpm install
+```
+
+### Setup Environment Variables
+
+Setup your OpenAI API key in the `.env.local` file:
+
+```bash
+OPENAI_API_KEY="" # setup your OpenAI API key here
+```
+
+### Run the Example
+
+```bash
+pnpm start
+```
+
+## Example
+
+The following example demonstrates how to build a handoff workflow:
+
+```typescript
+import assert from "node:assert";
+import { AIAgent, ChatModelOpenAI, ExecutionEngine } from "@aigne/core-next";
+
+const { OPENAI_API_KEY } = process.env;
+assert(OPENAI_API_KEY, "Please set the OPENAI_API_KEY environment variable");
+
+const model = new ChatModelOpenAI({
+  apiKey: OPENAI_API_KEY,
+});
+
+function transfer_to_b() {
+  return agentB;
+}
+
+const agentA = AIAgent.from({
+  name: "AgentA",
+  instructions: "You are a helpful agent.",
+  outputKey: "A",
+  tools: [transfer_to_b],
+});
+
+const agentB = AIAgent.from({
+  name: "AgentB",
+  instructions: "Only speak in Haikus.",
+  outputKey: "B",
+});
+
+const engine = new ExecutionEngine({ model });
+
+const userAgent = await engine.run(agentA);
+
+const result1 = await userAgent.call("transfer to agent b");
+console.log(result1);
+// Output:
+// {
+//   B: "Transfer now complete,  \nAgent B is here to help.  \nWhat do you need, friend?",
+// }
+
+const result2 = await userAgent.call("It's a beautiful day");
+console.log(result2);
+// Output:
+// {
+//   B: "Sunshine warms the earth,  \nGentle breeze whispers softly,  \nNature sings with joy.  ",
+// }
+```
+
+## License
+
+This project is licensed under the MIT License.

package/index.ts

@@ -0,0 +1,166 @@
+#!/usr/bin/env npx -y bun
+
+import assert from "node:assert";
+import {
+  AIAgent,
+  type Agent,
+  ChatModelOpenAI,
+  ExecutionEngine,
+  FunctionAgent,
+  runChatLoopInTerminal,
+} from "@aigne/core-next";
+import { z } from "zod";
+
+const { OPENAI_API_KEY } = process.env;
+assert(OPENAI_API_KEY, "Please set the OPENAI_API_KEY environment variable");
+
+const model = new ChatModelOpenAI({
+  apiKey: OPENAI_API_KEY,
+});
+
+const execute_order_tool = FunctionAgent.from({
+  name: "execute_order",
+  description: "Price should be in USD.",
+  inputSchema: z.object({
+    product: z.string(),
+    price: z.number(),
+  }),
+  fn: ({ product, price }: { product: string; price: number }) => {
+    console.log("\n\n=== Order Summary ===");
+    console.log(`Product: ${product}`);
+    console.log(`Price: $${price}`);
+    console.log("=================\n");
+    const confirm = "y";
+    if (confirm === "y") {
+      console.log("Order execution successful!");
+      return { result: "Success" };
+    }
+    console.log("Order cancelled!");
+    return { result: "User cancelled order." };
+  },
+});
+
+const look_up_item_tool = FunctionAgent.from({
+  name: "look_up_item",
+  description: "Use to find item ID.\nSearch query can be a description or keywords.",
+  inputSchema: z.object({
+    search_query: z.string(),
+  }),
+  fn: ({ search_query }: { search_query: string }) => {
+    const item_id = "item_132612938";
+    console.log(`Found item for: ${search_query}`, item_id);
+    return { item_id };
+  },
+});
+
+const execute_refund_tool = FunctionAgent.from({
+  name: "execute_refund",
+  description: "Use to execute a refund.",
+  inputSchema: z.object({
+    item_id: z.string(),
+    reason: z.string().optional(),
+  }),
+  fn: ({ item_id, reason }: { item_id: string; reason?: string }) => {
+    console.log("\n\n=== Refund Summary ===");
+    console.log(`Item ID: ${item_id}`);
+    console.log(`Reason: ${reason ?? "not provided"}`);
+    console.log("=================\n");
+    console.log("Refund execution successful!");
+    return { result: "success" };
+  },
+});
+
+const transfer_to_issues_and_repairs = FunctionAgent.from({
+  name: "transfer_to_issues_and_repairs",
+  description: "Use for issues, repairs, or refunds.",
+  fn: (): Agent => issuesAndRepairs,
+});
+
+const transfer_to_sales_agent = FunctionAgent.from({
+  name: "transfer_to_sales_agent",
+  description: "Use for anything sales or buying related.",
+  fn: (): Agent => sales,
+});
+
+const transfer_back_to_triage = FunctionAgent.from({
+  name: "transfer_back_to_triage",
+  description:
+    "Call this if the user brings up a topic outside of your purview,\nincluding escalating to human.",
+  fn: (): Agent => triage,
+});
+
+const transfer_to_human_manager = FunctionAgent.from({
+  name: "transfer_to_human_manager",
+  description: "Only call this if explicitly asked to.",
+  fn: (): Agent => humanAgent,
+});
+
+const sales = AIAgent.from({
+  name: "sales",
+  instructions: `\
+You are a sales agent for ACME Inc.
+Always answer in a sentence or less.
+Follow the following routine with the user:
+1. Ask them about any problems in their life related to catching roadrunners.
+2. Casually mention one of ACME's crazy made-up products can help.
+ - Don't mention price.
+3. Once the user is bought in, drop a ridiculous price.
+4. Only after everything, and if the user says yes,
+tell them a crazy caveat and execute their order.
+`,
+  tools: [transfer_back_to_triage, execute_order_tool],
+  outputKey: "sales",
+  enableHistory: true,
+});
+
+const issuesAndRepairs = AIAgent.from({
+  name: "issuesAndRepairs",
+  instructions: `\
+You are a customer support agent for ACME Inc.
+Always answer in a sentence or less.
+Follow the following routine with the user:
+1. First, ask probing questions and understand the user's problem deeper.
+ - unless the user has already provided a reason.
+2. Propose a fix (make one up).
+3. ONLY if not satisfied, offer a refund.
+4. If accepted, search for the ID and then execute refund.
+`,
+  tools: [transfer_back_to_triage, execute_refund_tool, look_up_item_tool],
+  outputKey: "issuesAndRepairs",
+  enableHistory: true,
+});
+
+// Assume this is a human agent
+const humanAgent = AIAgent.from({
+  name: "human_manager",
+  instructions: `\
+You are a human manager for ACME Inc.
+Just chat with the user and help them with their problem.
+Only transfer to another agent if user explicitly asks for it.
+`,
+  tools: [transfer_back_to_triage, transfer_to_sales_agent, transfer_to_issues_and_repairs],
+  outputKey: "human",
+  enableHistory: true,
+});
+
+const triage = AIAgent.from({
+  name: "triage",
+  instructions: `\
+You are a customer service bot for ACME Inc.
+Introduce yourself. Always be very brief.
+Gather information to direct the customer to the right department.
+But make your questions subtle and natural.
+`,
+  tools: [transfer_to_issues_and_repairs, transfer_to_sales_agent, transfer_to_human_manager],
+  outputKey: "triage",
+  enableHistory: true,
+});
+
+const engine = new ExecutionEngine({ model });
+
+const userAgent = await engine.run(triage);
+
+await runChatLoopInTerminal(userAgent, {
+  welcome: `Hello, I'm a customer service bot for ACME Inc. How can I help you today?`,
+  defaultQuestion: "I want a refund",
+});

package/package.json

@@ -0,0 +1,25 @@
+{
+  "name": "@aigne/example-workflow-handoff",
+  "version": "1.1.0-beta.12",
+  "description": "A demonstration of using AIGNE Framework to build a handoff workflow",
+  "author": "Arcblock <blocklet@arcblock.io> https://github.com/blocklet",
+  "homepage": "https://github.com/AIGNE-io/aigne-framework/tree/main/examples/workflow-handoff",
+  "license": "ISC",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/AIGNE-io/aigne-framework"
+  },
+  "bin": "index.ts",
+  "files": [
+    ".env.local.example",
+    "*.ts",
+    "README.md"
+  ],
+  "dependencies": {
+    "zod": "^3.24.2",
+    "@aigne/core-next": "^1.1.0-beta.12"
+  },
+  "scripts": {
+    "start": "npx -y bun run index.ts"
+  }
+}

package/usages.ts

@@ -0,0 +1,44 @@
+import assert from "node:assert";
+import { AIAgent, ChatModelOpenAI, ExecutionEngine } from "@aigne/core-next";
+
+const { OPENAI_API_KEY } = process.env;
+assert(OPENAI_API_KEY, "Please set the OPENAI_API_KEY environment variable");
+
+const model = new ChatModelOpenAI({
+  apiKey: OPENAI_API_KEY,
+});
+
+function transfer_to_b() {
+  return agentB;
+}
+
+const agentA = AIAgent.from({
+  name: "AgentA",
+  instructions: "You are a helpful agent.",
+  outputKey: "A",
+  tools: [transfer_to_b],
+});
+
+const agentB = AIAgent.from({
+  name: "AgentB",
+  instructions: "Only speak in Haikus.",
+  outputKey: "B",
+});
+
+const engine = new ExecutionEngine({ model });
+
+const userAgent = await engine.run(agentA);
+
+const result1 = await userAgent.call("transfer to agent b");
+console.log(result1);
+// Output:
+// {
+//   B: "Transfer now complete,  \nAgent B is here to help.  \nWhat do you need, friend?",
+// }
+
+const result2 = await userAgent.call("It's a beautiful day");
+console.log(result2);
+// Output:
+// {
+//   B: "Sunshine warms the earth,  \nGentle breeze whispers softly,  \nNature sings with joy.  ",
+// }