Hermes Agent开发实战：从API调用到Tool Use

from openai import OpenAI
client = OpenAI()

response = client.responses.create(
    model="gpt-5.4",
    input="Write a one-sentence bedtime story about a unicorn."
)

print(response.output_text)

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1000,
    messages=[
        {
            "role": "user",
            "content": "What should I search for to find the latest developments in renewable energy?",
        }
    ],
)
print(message.content)

import os
from openai import OpenAI

client = OpenAI(
    api_key='OPENAI_API_KEY',
    base_url='OPENAI_BASE_URL'
)
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "You are a helpful assistant"},
        {"role": "user", "content": "Hello"},
    ],
    stream=False
)

print(response.choices[0].message.content)

from openai import OpenAI

# 指向 LiteLLM 代理，一个入口调所有模型
client = OpenAI(
    api_key="API_KEY",
    base_url="BASE_URL",
)

# 只需切换 model 参数
for model in ["gpt-5", "claude-sonnet-4.6", "deepseek-v3.2"]:
    response = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": "用一句话介绍你自己。"}],
    )
    print(f"[{model}] {response.choices[0].message.content}")

# 第三轮请求时，messages 长这样：
messages = [
    {"role": "user", "content": "我想去日本旅行，有什么推荐？"},
    {"role": "assistant", "content": "推荐三个方向：1. 东京+镰仓（都市+海岸）2. 京都+大阪（古都+美食）3. 北海道（自然+温泉）"},
    {"role": "user", "content": "我比较喜欢第二个，大概需要几天？"},
    {"role": "assistant", "content": "京都+大阪建议 5-7 天。京都 3 天看寺庙和岚山，大阪 2 天吃道顿堀逛心斋桥。"},
    {"role": "user", "content": "帮我列一个 5 天的行程大纲。"},
]

response = client.messages.create(
    model="claude-sonnet-4.6",
    max_tokens=1000,
    system="你是一个旅行规划助手，帮用户规划旅行行程。回答简洁。",
    messages=messages,
)

用户提问 → 模型返回 tool_use（我要调这个函数） 
         → 开发者执行函数 
         → 把结果通过 tool_result 返回给模型 
         → 模型生成最终回答

import json
import os

import anthropic
import requests

client = anthropic.Anthropic()
AMAP_KEY = os.environ["AMAP_KEY"]

# ---- 第一步：定义工具 ----
# 告诉模型有哪些函数可用，传入函数名、描述和参数 schema
tools = [
    {
        "name": "get_weather",
        "description": "获取指定城市的当前天气信息",
        "input_schema": {
            "type": "object",
            "properties": {
                "city": {
                    "type": "string",
                    "description": "城市名称，如 北京、上海",
                },
            },
            "required": ["city"],
        },
    }
]

# ---- 第二步：实现工具 ----
# 这里调用高德天气 API 实现
def city_to_adcode(city: str) -> str | None:
    resp = requests.get(
        "https://restapi.amap.com/v3/geocode/geo",
        params={"key": AMAP_KEY, "address": city},
        timeout=5,
    )
    data = resp.json()
    if data.get("status") == "1" and data.get("geocodes"):
        return data["geocodes"][0]["adcode"]
    return None

def get_weather(city: str) -> dict:
    adcode = city_to_adcode(city)
    if not adcode:
        return {"error": f"无法识别城市「{city}」"}

    resp = requests.get(
        "https://restapi.amap.com/v3/weather/weatherInfo",
        params={"key": AMAP_KEY, "city": adcode, "extensions": "base"},
        timeout=5,
    )
    data = resp.json()
    live = data["lives"][0]
    return {
        "city": live["city"],
        "weather": live["weather"],
        "temperature": f"{live['temperature']}°C",
        "humidity": f"{live['humidity']}%",
        "wind": f"{live['winddirection']}风 {live['windpower']}级",
    }

# 工具注册表：动态分派，新增工具只需加一行映射
tool_registry = {
    "get_weather": get_weather,
}

# ---- 第三步：发起请求、执行工具、返回结果 ----
messages = [
    {"role": "user", "content": "北京和上海今天天气怎么样？"},
]

# 第一轮：发送用户消息，模型返回 tool_use
response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1000,
    tools=tools,
    messages=messages,
)

# stop_reason 是 "tool_use" 而非 "end_turn"
# 模型在说"我还没完，先帮我执行这些工具"
print(response.stop_reason)  # "tool_use"

# 把 assistant 的完整回复追加到对话历史
messages.append({"role": "assistant", "content": response.content})

# 执行工具：遍历 tool_use block，调用真实函数
tool_results = []
for block in response.content:
    if block.type != "tool_use":
        continue
    func = tool_registry[block.name]
    result = func(**block.input)
    tool_results.append({
        "type": "tool_result",
        "tool_use_id": block.id,
        "content": json.dumps(result, ensure_ascii=False),
    })

# 工具结果通过 user 角色的 tool_result 块返回
messages.append({"role": "user", "content": tool_results})

# 第二轮：带上完整历史，模型生成最终回答
final_response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1000,
    tools=tools,
    messages=messages,
)

print(final_response.content[0].text)

🏙️ 北京 — 雾，10°C，湿度 88%，东北风 ≤3级
🏙️ 上海 — 阴，26°C，湿度 56%，西风 ≤3级

import anthropic

client = anthropic.Anthropic()

tools = [
    {
        "type": "web_search_20260209",
        "name": "web_search",
        "max_uses": 3,
        "user_location": {
            "type": "approximate",
            "country": "CN",
            "city": "Beijing",
            "timezone": "Asia/Shanghai",
        },
    }
]

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=2000,
    tools=tools,
    messages=[{"role": "user", "content": "Claude Code 最新版本有什么新功能？"}],
)

# 响应中包含 web_search_tool_result 和最终文本
for block in response.content:
    if block.type == "text":
        print(block.text)

API 基础调用 → 多轮对话 → Tool Use → 内置工具