@langgraph-js/pro/README.md

The Pro SDK for LangGraph - seamlessly integrate your AI agents with frontend interfaces and build complex AI workflows

# @langgraph-js/pro

![npm version](https://img.shields.io/npm/v/@langgraph-js/pro)
![license](https://img.shields.io/npm/l/@langgraph-js/pro)

This is an extension library for LangGraph that provides additional convenient features and tools, helping developers build complex AI workflows more easily. It's particularly suitable for building multi-agent collaboration, deep research, task planning, and other complex scenarios.

> LangGraph 1.0 Support!

## Getting Started

### 1. State Management

```typescript
import { AgentState } from "@langgraph-js/pro";
import { z } from "zod";
// 创建自定义状态
const PlannerState = AgentState.merge(
    z.object({
        current_plan: z.any().nullable(),
        title: z.string(),
    })
);

// 扩展已有状态
const ExportState = PlannerState.merge(
    z.object({
        description: z.string(),
    })
);
```

### 2. Swarm Agent

```typescript
import { createAgent } from "langchain";
import { createSwarm, createHandoffTool } from "@langgraph-js/pro";

// Create coordinator agent
const coordinator_agent = createAgent({
    name: "coordinator",
    model,
    tools: [
        createHandoffTool({
            agentName: "planner",
            description: "Create plan",
            // it will keep all state sync with Swarm
            updateState: keepAllStateInHandOff,
        }),
    ],
    systemPrompt: "",
});

// Create swarm collaboration
const swarm = createSwarm({
    agents: [coordinator, planner, researcher, reporter],
    defaultActiveAgent: "coordinator",
});
```

### 3. Frontend Actions (Tools) Support

```typescript
import { createFeTools, FEToolsState } from "@langgraph-js/pro";

// Define tools
const tools = [
    {
        name: "search",
        description: "Search information",
        parameters: [
            {
                name: "query",
                type: "string",
                description: "Search keywords",
                required: true,
            },
        ],
    },
];

// Create state with tools
const State = createState(FEToolsState).build({});

// Use tools in node
const node = (state) => {
    const feTools = createFeTools(state.fe_tools);
    llm.bindTools(feTools);
};
```

### 4. Messages Utils

```typescript
import { getLastHumanMessage, getTextMessageContent } from "@langgraph-js/pro";

// Get last human message
const lastMessage = getLastHumanMessage(state.messages);

// Get message text content
const content = getTextMessageContent(message);
```

### 5. Model Helper

```typescript
import { createModelHelper } from "@langgraph-js/pro";

// 创建模型状态和工具
const { ModelState, createLLM } = createModelHelper({
    main_model: ["gemini-2.5-flash"],
    assistant_model: ["gpt-4-turbo-preview"],
});

// 扩展状态
const GraphState = createState(ModelState).build({});

// 使用模型
const llm = await createLLM(state, "main_model", {
    temperature: 0.7,
    streaming: true,
});
```

### 6. Web Tools

```typescript
import { crawler_tool, web_search_tool } from "@langgraph-js/pro";

//  env file SEARCH_SERVER_URL=https://website-to-md.deno.dev
const tools = [crawler_tool, web_search_tool];
```

## License

This project is licensed under the Apache-2.0 License.