# SDK

TypeScript 和 Python 的类型化客户端，都从 AREX_API_KEY 读取密钥。

## TypeScript

research.run 创建 Run 并等待答案；research.create、get、wait、cancel 和 delete 可分步操作。错误均为 ArexError 的子类。

### Shell

```bash
npm install @arex-research/ts-sdk
```

### TypeScript

```typescript
import { Arex } from "@arex-research/ts-sdk";

const arex = new Arex(); // reads AREX_API_KEY

const { results } = await arex.search("how are AI agents evaluated", { limit: 5 });
for (const { url } of results) {
  const page = await arex.contents(url, { maxCharacters: 4000 });
}
const run = await arex.research.run("Compare agent evaluation benchmarks", { mode: "fast" });
console.log(run.result.answer);
```

## Python

需要 Python 3.10 及以上。AsyncArex 提供相同的异步调用。

### Shell

```bash
pip install arex-python-sdk
```

### Python

```python
from arex import Arex

with Arex() as arex:  # reads AREX_API_KEY
    search = arex.search("how are AI agents evaluated", limit=5)
    for item in search.results:
        page = arex.contents(item.url, max_characters=4000)
    run = arex.research.run("Compare agent evaluation benchmarks", mode="fast")
    assert run.result is not None  # always set once run() returns
    print(run.result.answer)
```

## 作为 Agent 工具

在 Pi SDK 或 LangGraph 中把 SDK 调用封装成工具。

### Pi SDK

```typescript
import { Arex } from "@arex-research/ts-sdk";

const arex = new Arex(); // reads AREX_API_KEY

export const arexTools = [
  {
    name: "arex_search",
    description: "Search the web for ranked URLs and snippets.",
    parameters: {
      type: "object",
      properties: { query: { type: "string" }, limit: { type: "number" } },
      required: ["query"],
    },
    execute: async ({ query, limit = 5 }: { query: string; limit?: number }) =>
      arex.search(query, { limit }),
  },
  {
    name: "arex_contents",
    description: "Read clean markdown text from a public URL.",
    parameters: {
      type: "object",
      properties: { url: { type: "string" } },
      required: ["url"],
    },
    execute: async ({ url }: { url: string }) =>
      arex.contents(url, { maxCharacters: 8000 }),
  },
  {
    name: "arex_research",
    description: "Answer a question with cited web sources in one call.",
    parameters: {
      type: "object",
      properties: { query: { type: "string" } },
      required: ["query"],
    },
    execute: async ({ query }: { query: string }) =>
      arex.research.run(query, { mode: "fast" }),
  },
];
```

### LangGraph

```python
from dataclasses import asdict
from arex import Arex
from langchain_core.tools import tool
from langgraph.prebuilt import create_react_agent

client = Arex()  # reads AREX_API_KEY

@tool
def arex_search(query: str, limit: int = 5) -> dict:
    """Search the web for ranked URLs and snippets."""
    return asdict(client.search(query, limit=limit))

@tool
def arex_contents(url: str) -> str:
    """Read clean markdown text from a public URL."""
    return client.contents(url, max_characters=8000).content

@tool
def arex_research(query: str) -> str:
    """Answer a research question with cited sources in one call."""
    run = client.research.run(query, mode="fast")
    return run.result.answer if run.result else ""

agent = create_react_agent("openai:gpt-4.1", tools=[arex_search, arex_contents, arex_research])
```

[npm · @arex-research/ts-sdk](https://www.npmjs.com/package/@arex-research/ts-sdk)

[PyPI · arex-python-sdk](https://pypi.org/project/arex-python-sdk/)

[GitHub](https://github.com/AREX-Research/arex-sdk)

[TypeScript Changelog](https://github.com/AREX-Research/arex-sdk/blob/main/ts/CHANGELOG.md)

[Python Changelog](https://github.com/AREX-Research/arex-sdk/blob/main/python/CHANGELOG.md)