一文徹底搞懂智能體Agent基于ReAct的工具調用

作者：AI大模型應用開發 2025-06-27 06:00:00

AI智能體是指具備一定自主性、能感知環境并通過智能決策執行特定任務的軟件或硬件實體。它結合了人工智能技術，能夠獨立或協作完成目標。本文包含ReAct基礎，基于ReAct的工具調用流程和涉及的交互消息，手搓Agent代碼實現基于ReAct的工具調用。

前言

AI智能體是指具備一定自主性、能感知環境并通過智能決策執行特定任務的軟件或硬件實體。它結合了人工智能技術（如機器學習、自然語言處理、計算機視覺等），能夠獨立或協作完成目標。

基于大語言模型（LLM）的Function Calling可以令智能體實現有效的工具使用和與外部API的交互。支持Function Calling的模型(如gpt-4，qwen-plus等)能夠檢測何時需要調用函數，并輸出調用函數的函數名和所需參數的JSON格式結構化數據。

但并非所有的LLM模型都支持Function Calling(如deepseek-v3)。對于不支持Function Calling的模型，可通過ReAct的相對較為復雜的提示詞工程，要求模型返回特定格式的響應，以便區分不同的階段（思考、行動、觀察）。

工具調用主要有兩個用途：

獲取數據: 例如根據關鍵字從知識庫檢索內容、通過特定API接口獲取業務數據
執行行動: 例如通過API接口修改業務狀態數據、執行預定業務操作

本文包含如下內容：

ReAct基礎
詳細介紹基于ReAct的工具調用流程和涉及的交互消息
手搓Agent代碼實現基于ReAct的工具調用

ReAct基礎

ReAct源于經典論文: REACT: SYNERGIZING REASONING AND ACTING IN LANGUAGE MODELS (鏈接：https://arxiv.org/pdf/2210.03629)

基于ReAct的智能體為了解決問題，需要經過幾個階段

Thought: 思考推理
Action：作出行動，決定要調用的工具和參數
Observation：行動的結果(工具輸出)

以上3個階段可能迭代多次，直到問題得到解決或者達到迭代次數上限。

基于ReAct的工具調用依賴于復雜的提示詞工程。系統提示詞參考langchain的模板:

Answer the following questions as best you can. You have access to the following tools:
{tool_strings}


The way you use the tools is by specifying a json blob.
Specifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).


The only values that should be in the "action" field are: {tool_names}


The $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:


```
{{{{
"action": $TOOL_NAME,
"action_input": $INPUT
}}}}
```


ALWAYS use the following format:


Question: the input question you must answer
Thought: you should always think about what to do
Action:
```
$JSON_BLOB
```
Observation: the result of the action
... (this Thought/Action/Observation can repeat N times)
Thought: I now know the final answer
Final Answer: the final answer to the original input question


Begin! Reminder to always use the exact characters `Final Answer` when responding.

基于ReAct的工具調用流程和交互消息

我們以查詢北京和廣州天氣為例，LLM采用阿里云的deepseek-v3。查詢天氣的流程如下圖：

1. 發起查詢請求

向LLM發起查詢時，messages列表有2條messages：

第1條role為system，定義了系統提示詞(含工具定義)
第2條role為user，包含如下內容：

Question: 北京和廣州天氣怎么樣

我們用curl發起POST請求，body的JSON結構可參考https://platform.openai.com/docs/api-reference/chat/create 。

請求里的stop字段需要設置為Observation:，否則LLM會直接輸出整個Thought/Action/Observation流程并給出虛構的最終答案。我們僅需要LLM輸出Thought/Action即可

#!/bin/bash
export OPENAI_BASE_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"
export OPENAI_API_KEY="sk-xxx" # 替換為你的key


curl ${OPENAI_BASE_URL}/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
  "model": "deepseek-v3",
  "messages": [
    {
      "role": "system",
      "content": "\nAnswer the following questions as best you can. You have access to the following tools:\n{\"name\": \"get_weather\", \"description\": \"Get weather\", \"parameters\": {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"the name of the location\"}}, \"required\": [\"location\"]}}\n\n\nThe way you use the tools is by specifying a json blob.\nSpecifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).\n\nThe only values that should be in the \"action\" field are: get_weather\n\nThe $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:\n\n```\n{{\n\"action\": $TOOL_NAME,\n\"action_input\": $INPUT\n}}\n```\n\nALWAYS use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction:\n```\n$JSON_BLOB\n```\nObservation: the result of the action\n... (this Thought/Action/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\n\nBegin! Reminder to always use the exact characters `Final Answer` when responding. \n"
    },
    {
      "role": "user",
      "content": "Question: 北京和廣州天氣怎么樣\n\n"
    }
  ],
  "stop": "Observation:"
}'

2. LLM返回Action獲取北京天氣

LLM經過推理，發現需要先調用函數獲取北京天氣。

Thought: 我需要獲取北京和廣州的天氣信息。首先，我將獲取北京的天氣。
Action:
```
{
  "action": "get_weather",
  "action_input": {
    "location": "北京"
  }
}
```

完整的JSON響應如下：

{
  "choices": [
    {
      "message": {
        "content": "Thought: 我需要獲取北京和廣州的天氣信息。首先，我將獲取北京的天氣。\n\nAction:\n```\n{\n  \"action\": \"get_weather\",\n  \"action_input\": {\n    \"location\": \"北京\"\n  }\n}\n```",
        "role": "assistant"
      },
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null
    }
  ],
  "object": "chat.completion",
  "usage": {
    "prompt_tokens": 305,
    "completion_tokens": 49,
    "total_tokens": 354
  },
  "created": 1745651748,
  "system_fingerprint": null,
  "model": "deepseek-v3",
  "id": "chatcmpl-697b0627-4fca-975b-954c-7304386ac224"
}

3. 處理函數調用獲取北京天氣

解析處理LLM的Action獲得函數名和參數列表，調用相應的API接口獲得結果。

例如：通過http://weather.cma.cn/api/now/54511可獲得北京的天氣情況。

完整的JSON響應如下：

{
  "msg": "success",
  "code": 0,
  "data": {
    "location": {
      "id": "54511",
      "name": "北京",
      "path": "中國, 北京, 北京"
    },
    "now": {
      "precipitation": 0.0,
      "temperature": 23.4,
      "pressure": 1005.0,
      "humidity": 43.0,
      "windDirection": "西南風",
      "windDirectionDegree": 216.0,
      "windSpeed": 2.7,
      "windScale": "微風",
      "feelst": 23.1
    },
    "alarm": [],
    "jieQi": "",
    "lastUpdate": "2025/04/26 15:00"
  }
}

4. 把上下文信息以及函數調用結果發給LLM

發給LLM的messages列表有2條messages：

第1條role為system，定義了系統提示詞(含工具定義)
第2條role為user，包含如下內容：

Question: 北京和廣州天氣怎么樣

Thought: 我需要獲取北京和廣州的天氣信息。首先，我將獲取北京的天氣

Action: {"action":"get_weather","action_input":{"location":"北京"}}

Observation: 工具調用get_weather('北京')的結果

#!/bin/bash
export OPENAI_BASE_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"
export OPENAI_API_KEY="sk-xxx" # 替換為你的key


curl ${OPENAI_BASE_URL}/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
  "model": "deepseek-v3",
  "messages": [
    {
      "role": "system",
      "content": "\nAnswer the following questions as best you can. You have access to the following tools:\n{\"name\": \"get_weather\", \"description\": \"Get weather\", \"parameters\": {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"the name of the location\"}}, \"required\": [\"location\"]}}\n\n\nThe way you use the tools is by specifying a json blob.\nSpecifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).\n\nThe only values that should be in the \"action\" field are: get_weather\n\nThe $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:\n\n```\n{{\n\"action\": $TOOL_NAME,\n\"action_input\": $INPUT\n}}\n```\n\nALWAYS use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction:\n```\n$JSON_BLOB\n```\nObservation: the result of the action\n... (this Thought/Action/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\n\nBegin! Reminder to always use the exact characters `Final Answer` when responding. \n"
    },
    {
      "role": "user",
      "content": "Question: 北京和廣州天氣怎么樣\n\nThought: 我需要獲取北京和廣州的天氣信息。首先，我將獲取北京的天氣。\n\nAction:\n```\n{\n\"action\": \"get_weather\",\n\"action_input\": {\n\"location\": \"北京\"\n}\n}\n```\nObservation: {\"msg\":\"success\",\"code\":0,\"data\":{\"location\":{\"id\":\"54511\",\"name\":\"北京\",\"path\":\"中國, 北京, 北京\"},\"now\":{\"precipitation\":0.0,\"temperature\":23.4,\"pressure\":1005.0,\"humidity\":43.0,\"windDirection\":\"西南風\",\"windDirectionDegree\":216.0,\"windSpeed\":2.7,\"windScale\":\"微風\",\"feelst\":23.1},\"alarm\":[],\"jieQi\":\"\",\"lastUpdate\":\"2025/04/26 15:00\"}}\n"
    }
  ],
  "stop": "Observation:"
}'

5. LLM返回Action獲取廣州天氣

LLM經過推理，發現還需要調用函數獲取廣州天氣。

Thought: 我已經獲取了北京的天氣信息。接下來，我將獲取廣州的天氣信息。
Action:
```
{
  "action": "get_weather",
  "action_input": {
    "location": "廣州"
  }
}
```

完整的JSON響應如下：

{
  "choices": [
    {
      "message": {
        "content": "Thought: 我已經獲取了北京的天氣信息。接下來，我將獲取廣州的天氣信息。\n\nAction:\n```\n{\n\"action\": \"get_weather\",\n\"action_input\": {\n\"location\": \"廣州\"\n}\n}\n```\nObservation",
        "role": "assistant"
      },
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null
    }
  ],
  "object": "chat.completion",
  "usage": {
    "prompt_tokens": 472,
    "completion_tokens": 46,
    "total_tokens": 518
  },
  "created": 1745651861,
  "system_fingerprint": null,
  "model": "deepseek-v3",
  "id": "chatcmpl-a822b8d7-9105-9dc2-8e98-4327afb50b3a"
}

6. 處理函數調用獲取廣州天氣

解析處理LLM的Action獲得函數名和參數列表，調用相應的API接口獲得結果。

例如：通過http://weather.cma.cn/api/now/59287可獲得廣州的天氣情況。

完整的JSON響應如下：

{
  "msg": "success",
  "code": 0,
  "data": {
    "location": {
      "id": "59287",
      "name": "廣州",
      "path": "中國, 廣東, 廣州"
    },
    "now": {
      "precipitation": 0.0,
      "temperature": 24.2,
      "pressure": 1005.0,
      "humidity": 79.0,
      "windDirection": "東北風",
      "windDirectionDegree": 31.0,
      "windSpeed": 1.3,
      "windScale": "微風",
      "feelst": 27.1
    },
    "alarm": [],
    "jieQi": "",
    "lastUpdate": "2025/04/26 15:00"
  }
}

7. 把上下文信息以及函數調用結果發給LLM

發給LLM的messages列表有2條messages：

第1條role為system，定義了系統提示詞(含工具定義)
第2條role為user，包含如下內容：

Question: 北京和廣州天氣怎么樣

Thought: 我需要獲取北京和廣州的天氣信息。首先，我將獲取北京的天氣

Action: {"action":"get_weather","action_input":{"location":"北京"}}

Observation: 工具調用get_weather('北京')的結果

Thought: 現在我已經獲取了北京的天氣信息，接下來我將獲取廣州的天氣信息。

Action: {"action":"get_weather","action_input":{"location":"廣州"}}

Observation: 工具調用get_weather('廣州')的結果

#!/bin/bash
export OPENAI_BASE_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"
export OPENAI_API_KEY="sk-xxx" # 替換為你的key


curl ${OPENAI_BASE_URL}/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
  "model": "deepseek-v3",
  "messages": [
    {
      "role": "system",
      "content": "\nAnswer the following questions as best you can. You have access to the following tools:\n{\"name\": \"get_weather\", \"description\": \"Get weather\", \"parameters\": {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"the name of the location\"}}, \"required\": [\"location\"]}}\n\n\nThe way you use the tools is by specifying a json blob.\nSpecifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).\n\nThe only values that should be in the \"action\" field are: get_weather\n\nThe $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:\n\n```\n{{\n\"action\": $TOOL_NAME,\n\"action_input\": $INPUT\n}}\n```\n\nALWAYS use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction:\n```\n$JSON_BLOB\n```\nObservation: the result of the action\n... (this Thought/Action/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\n\nBegin! Reminder to always use the exact characters `Final Answer` when responding. \n"
    },
    {
      "role": "user",
      "content": "Question: 北京和廣州天氣怎么樣\n\nThought: 我需要獲取北京和廣州的天氣信息。首先，我將獲取北京的天氣。\n\nAction:\n```\n{\n\"action\": \"get_weather\",\n\"action_input\": {\n\"location\": \"北京\"\n}\n}\n```\nObservation: {\"msg\":\"success\",\"code\":0,\"data\":{\"location\":{\"id\":\"54511\",\"name\":\"北京\",\"path\":\"中國, 北京, 北京\"},\"now\":{\"precipitation\":0.0,\"temperature\":23.4,\"pressure\":1005.0,\"humidity\":43.0,\"windDirection\":\"西南風\",\"windDirectionDegree\":216.0,\"windSpeed\":2.7,\"windScale\":\"微風\",\"feelst\":23.1},\"alarm\":[],\"jieQi\":\"\",\"lastUpdate\":\"2025/04/26 15:00\"}}\nThought: 現在我已經獲取了北京的天氣信息，接下來我將獲取廣州的天氣信息。\n\nAction:\n```\n{\n\"action\": \"get_weather\",\n\"action_input\": {\n\"location\": \"廣州\"\n}\n}\n```\nObservation\nObservation: {\"msg\":\"success\",\"code\":0,\"data\":{\"location\":{\"id\":\"59287\",\"name\":\"廣州\",\"path\":\"中國, 廣東, 廣州\"},\"now\":{\"precipitation\":0.0,\"temperature\":24.2,\"pressure\":1005.0,\"humidity\":79.0,\"windDirection\":\"東北風\",\"windDirectionDegree\":31.0,\"windSpeed\":1.3,\"windScale\":\"微風\",\"feelst\":27.1},\"alarm\":[],\"jieQi\":\"\",\"lastUpdate\":\"2025/04/26 15:00\"}}\n"
    }
  ],
  "stop": "Observation:"
}'

8. LLM生成最終回復

LLM生成最終的回復：

Thought: 我已經獲取了北京和廣州的天氣信息，現在可以回答用戶的問題了。


Final Answer: 北京的天氣溫度為23.4°C，濕度為43%，風向為西南風，風速為2.7米/秒。廣州 的天氣溫度為24.2°C，濕度為79%，風向為東北風，風速為1.3米/秒。

完整的JSON響應如下：

{
  "choices": [
    {
      "message": {
        "content": "Thought: 我已經獲取了北京和廣州的天氣信息，現在可以回答用戶的問題了。\n\nFinal Answer: 北京的天氣溫度為23.4°C，濕度為43%，風向為西南風，風速為2.7米/秒。廣州 的天氣溫度為24.2°C，濕度為79%，風向為東北風，風速為1.3米/秒。",
        "role": "assistant"
      },
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null
    }
  ],
  "object": "chat.completion",
  "usage": {
    "prompt_tokens": 641,
    "completion_tokens": 79,
    "total_tokens": 720
  },
  "created": 1745652025,
  "system_fingerprint": null,
  "model": "deepseek-v3",
  "id": "chatcmpl-d9b85f31-589e-9c6f-8694-cf813344e464"
}

手搓Agent代碼實現基于ReAct的工具調用

1. 創建python環境

uv init agent
cd agent
uv venv
.venv\Scripts\activate

uv add openai requests python-dotenv

2. 設置API Key

創建.env，.env內容如下（注意修改OPENAI_API_KEY為您的key）

OPENAI_API_KEY=your_api_key_here
OPENAI_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1

把.env添加到.gitignore

3. 實現Agent代碼

基于openai sdk實現ReAct agent的偽代碼主體邏輯如下：

maxIter = 5 # 最大迭代次數
agent_scratchpad = "" # agent思考過程(Thought/Action/Observation)
for iterSeq in range(1, maxIter+1):
    構造chat completion請求
        messages有2條
            第1條為系統提示詞消息(含工具定義)
            第2條為用戶消息：Question + agent思考過程(Thought/Action/Observation)
        stop參數設置為"Observation:"
    獲取chat completion結果
    如果chat completion結果帶有"Final Answer:"
        返回最終答案
    如果chat completion結果帶有Action
        解析并調用相應函數
        更新agent思考過程：把本次LLM的輸出(Though/Action)和工具調用結果(Observation)添加到agent_scratchpad
        繼續迭代

完整的main.py代碼如下：

import json
import re
import requests
import urllib.parse
from typing import Iterable
from openai import OpenAI
from openai.types.chat.chat_completion_message_param import ChatCompletionMessageParam
from openai.types.chat.chat_completion_user_message_param import (
    ChatCompletionUserMessageParam,
)
from openai.types.chat.chat_completion_system_message_param import (
    ChatCompletionSystemMessageParam,
)


# 加載環境變量
from dotenv import load_dotenv
load_dotenv()


client = OpenAI()
model = "deepseek-v3"


# 工具定義
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get weather",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "location"}
                },
                "required": ["location"],
            },
        },
    }
]


# 系統提示詞
def get_system_prompt():
    tool_strings = "\n".join([json.dumps(tool["function"]) for tool in tools])
    tool_names = ", ".join([tool["function"]["name"] for tool in tools])
    systemPromptFormat = """
Answer the following questions as best you can. You have access to the following tools:
{tool_strings}




The way you use the tools is by specifying a json blob.
Specifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).


The only values that should be in the "action" field are: {tool_names}


The $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:


```
{{{{
"action": $TOOL_NAME,
"action_input": $INPUT
}}}}
```


ALWAYS use the following format:


Question: the input question you must answer
Thought: you should always think about what to do
Action:
```
$JSON_BLOB
```
Observation: the result of the action
... (this Thought/Action/Observation can repeat N times)
Thought: I now know the final answer
Final Answer: the final answer to the original input question




Begin! Reminder to always use the exact characters `Final Answer` when responding. 
"""
    return systemPromptFormat.format(tool_strings=tool_strings, tool_names=tool_names)


# 實現獲取天氣
def get_weather(location: str) -> str:
    url = "http://weather.cma.cn/api/autocomplete?q=" + urllib.parse.quote(location)
    response = requests.get(url)
    data = response.json()
    if data["code"] != 0:
        return "沒找到該位置的信息"
    location_code = ""
    for item in data["data"]:
        str_array = item.split("|")
        if (
            str_array[1] == location
            or str_array[1] + "市" == location
            or str_array[2] == location
        ):
            location_code = str_array[0]
            break
    if location_code == "":
        return "沒找到該位置的信息"
    url = f"http://weather.cma.cn/api/now/{location_code}"
    return requests.get(url).text


# 實現工具調用
def invoke_tool(toolName: str, toolParamaters) -> str:
    result = ""
    if toolName == "get_weather":
        result = get_weather(toolParamaters["location"])
    else:
        result = f"函數{toolName}未定義"
    return result


def main():
    query = "北京和廣州天氣怎么樣"
    systemMsg = ChatCompletionSystemMessageParam(
        role="system", cnotallow=get_system_prompt()
    )
    maxIter = 5  # 最大迭代次數
    agent_scratchpad = ""  # agent思考過程
    action_pattern = re.compile(r"\nAction:\n`{3}(?:json)?\n(.*?)`{3}.*?$", re.DOTALL)
    for iterSeq in range(1, maxIter + 1):
        messages: Iterable[ChatCompletionMessageParam] = list()
        messages.append(systemMsg)
        messages.append(
            ChatCompletionUserMessageParam(
                role="user", cnotallow=f"Question: {query}\n\n{agent_scratchpad}"
            )
        )
        print(f">> iterSeq:{iterSeq}")
        print(f">>> messages: {json.dumps(messages)}")
        # 向LLM發起請求，注意需要設置stop參數
        chat_completion = client.chat.completions.create(
            messages=messages,
            model=model,
            stop="Observation:",
        )
        content = chat_completion.choices[0].message.content
        print(f">>> content:\n{content}")
        final_answer_match = re.search(r"\nFinal Answer:\s*(.*)", content)
        if final_answer_match:
            final_answer = final_answer_match.group(1)
            print(f">>> 最終答案: {final_answer}")
            return
        action_match = action_pattern.search(content)
        if action_match:
            obj = json.loads(action_match.group(1))
            toolName = obj["action"]
            toolParameters = obj["action_input"]
            print(f">>> tool name:{toolName}")
            print(f">>> tool parameters:{toolParameters}")
            result = invoke_tool(toolName, toolParameters)
            print(f">>> tool result: {result}")
            # 把本次LLM的輸出(Though/Action)和工具調用結果(Observation)添加到agent_scratchpad
            agent_scratchpad += content + f"\nObservation: {result}\n"
        else:
            print(">>> ERROR: detect invalid response")
            return
    print(">>> 迭代次數達到上限，我無法得到最終答案")


main()

運行代碼：uv run .\main.py

輸出日志如下：

>> iterSeq:1
>>> messages: [{"role": "system", "content": "\nAnswer the following questions as best you can. You have access to the following tools:\n{\"name\": \"get_weather\", \"description\": \"Get weather\", \"parameters\": {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"the name of the location\"}}, \"required\": [\"location\"]}}\n\n\nThe way you use the tools is by specifying a json blob.\nSpecifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).\n\nThe only values that should be in the \"action\" field are: get_weather\n\nThe $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:\n\n```\n{{\n\"action\": $TOOL_NAME,\n\"action_input\": $INPUT\n}}\n```\n\nALWAYS use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction:\n```\n$JSON_BLOB\n```\nObservation: the result of the action\n... (this Thought/Action/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\n\nBegin! Reminder to always use the exact characters `Final Answer` when responding. \n"}, {"role": "user", "content": "Question: \u5317\u4eac\u548c\u5e7f\u5dde\u5929\u6c14\u600e\u4e48\u6837\n\n"}]
>>> content:
Thought: 我需要獲取北京和廣州的天氣信息。首先，我將獲取北京的天氣。


Action:
```
{
"action": "get_weather",
"action_input": {
"location": "北京"
}
}
```
>>> tool name:get_weather
>>> tool parameters:{'location': '北京'}
>>> tool result: {"msg":"success","code":0,"data":{"location":{"id":"54511","name":"北京","path":"中國, 北京, 北京"},"now":{"precipitation":0.0,"temperature":23.4,"pressure":1005.0,"humidity":43.0,"windDirection":"西南風","windDirectionDegree":216.0,"windSpeed":2.7,"windScale":"微風","feelst":23.1},"alarm":[],"jieQi":"","lastUpdate":"2025/04/26 15:00"}}
>> iterSeq:2
>>> messages: [{"role": "system", "content": "\nAnswer the following questions as best you can. You have access to the following tools:\n{\"name\": \"get_weather\", \"description\": \"Get weather\", \"parameters\": {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"the name of the location\"}}, \"required\": [\"location\"]}}\n\n\nThe way you use the tools is by specifying a json blob.\nSpecifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).\n\nThe only values that should be in the \"action\" field are: get_weather\n\nThe $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:\n\n```\n{{\n\"action\": $TOOL_NAME,\n\"action_input\": $INPUT\n}}\n```\n\nALWAYS use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction:\n```\n$JSON_BLOB\n```\nObservation: the result of the action\n... (this Thought/Action/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\n\nBegin! Reminder to always use the exact characters `Final Answer` when responding. \n"}, {"role": "user", "content": "Question: \u5317\u4eac\u548c\u5e7f\u5dde\u5929\u6c14\u600e\u4e48\u6837\n\nThought: \u6211\u9700\u8981\u83b7\u53d6\u5317\u4eac\u548c\u5e7f\u5dde\u7684\u5929\u6c14\u4fe1\u606f\u3002\u9996\u5148\uff0c\u6211\u5c06\u83b7\u53d6\u5317\u4eac\u7684\u5929\u6c14\u3002\n\nAction:\n```\n{\n\"action\": \"get_weather\",\n\"action_input\": {\n\"location\": \"\u5317\u4eac\"\n}\n}\n```\nObservation: {\"msg\":\"success\",\"code\":0,\"data\":{\"location\":{\"id\":\"54511\",\"name\":\"\u5317\u4eac\",\"path\":\"\u4e2d\u56fd, \u5317\u4eac, \u5317\u4eac\"},\"now\":{\"precipitation\":0.0,\"temperature\":23.4,\"pressure\":1005.0,\"humidity\":43.0,\"windDirection\":\"\u897f\u5357\u98ce\",\"windDirectionDegree\":216.0,\"windSpeed\":2.7,\"windScale\":\"\u5fae\u98ce\",\"feelst\":23.1},\"alarm\":[],\"jieQi\":\"\",\"lastUpdate\":\"2025/04/26 15:00\"}}\n"}]
>>> content:
Thought: 現在我已經獲取了北京的天氣信息，接下來我將獲取廣州的天氣信息。


Action:
```
{
"action": "get_weather",
"action_input": {
"location": "廣州"
}
}
```
Observation
>>> tool name:get_weather
>>> tool parameters:{'location': '廣州'}
>>> tool result: {"msg":"success","code":0,"data":{"location":{"id":"59287","name":"廣州","path":"中國, 廣東, 廣州"},"now":{"precipitation":0.0,"temperature":24.2,"pressure":1005.0,"humidity":79.0,"windDirection":"東北風","windDirectionDegree":31.0,"windSpeed":1.3,"windScale":"微風","feelst":27.1},"alarm":[],"jieQi":"","lastUpdate":"2025/04/26 15:00"}}
>> iterSeq:3
>>> messages: [{"role": "system", "content": "\nAnswer the following questions as best you can. You have access to the following tools:\n{\"name\": \"get_weather\", \"description\": \"Get weather\", \"parameters\": {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"the name of the location\"}}, \"required\": [\"location\"]}}\n\n\nThe way you use the tools is by specifying a json blob.\nSpecifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).\n\nThe only values that should be in the \"action\" field are: get_weather\n\nThe $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:\n\n```\n{{\n\"action\": $TOOL_NAME,\n\"action_input\": $INPUT\n}}\n```\n\nALWAYS use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction:\n```\n$JSON_BLOB\n```\nObservation: the result of the action\n... (this Thought/Action/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\n\nBegin! Reminder to always use the exact characters `Final Answer` when responding. \n"}, {"role": "user", "content": "Question: \u5317\u4eac\u548c\u5e7f\u5dde\u5929\u6c14\u600e\u4e48\u6837\n\nThought: \u6211\u9700\u8981\u83b7\u53d6\u5317\u4eac\u548c\u5e7f\u5dde\u7684\u5929\u6c14\u4fe1\u606f\u3002\u9996\u5148\uff0c\u6211\u5c06\u83b7\u53d6\u5317\u4eac\u7684\u5929\u6c14\u3002\n\nAction:\n```\n{\n\"action\": \"get_weather\",\n\"action_input\": {\n\"location\": \"\u5317\u4eac\"\n}\n}\n```\nObservation: {\"msg\":\"success\",\"code\":0,\"data\":{\"location\":{\"id\":\"54511\",\"name\":\"\u5317\u4eac\",\"path\":\"\u4e2d\u56fd, \u5317\u4eac, \u5317\u4eac\"},\"now\":{\"precipitation\":0.0,\"temperature\":23.4,\"pressure\":1005.0,\"humidity\":43.0,\"windDirection\":\"\u897f\u5357\u98ce\",\"windDirectionDegree\":216.0,\"windSpeed\":2.7,\"windScale\":\"\u5fae\u98ce\",\"feelst\":23.1},\"alarm\":[],\"jieQi\":\"\",\"lastUpdate\":\"2025/04/26 15:00\"}}\nThought: \u73b0\u5728\u6211\u5df2\u7ecf\u83b7\u53d6\u4e86\u5317\u4eac\u7684\u5929\u6c14\u4fe1\u606f\uff0c\u63a5\u4e0b\u6765\u6211\u5c06\u83b7\u53d6\u5e7f\u5dde\u7684\u5929\u6c14\u4fe1\u606f\u3002\n\nAction:\n```\n{\n\"action\": \"get_weather\",\n\"action_input\": {\n\"location\": \"\u5e7f\u5dde\"\n}\n}\n```\nObservation\nObservation: {\"msg\":\"success\",\"code\":0,\"data\":{\"location\":{\"id\":\"59287\",\"name\":\"\u5e7f\u5dde\",\"path\":\"\u4e2d\u56fd, \u5e7f\u4e1c, \u5e7f\u5dde\"},\"now\":{\"precipitation\":0.0,\"temperature\":24.2,\"pressure\":1005.0,\"humidity\":79.0,\"windDirection\":\"\u4e1c\u5317\u98ce\",\"windDirectionDegree\":31.0,\"windSpeed\":1.3,\"windScale\":\"\u5fae\u98ce\",\"feelst\":27.1},\"alarm\":[],\"jieQi\":\"\",\"lastUpdate\":\"2025/04/26 15:00\"}}\n"}]
>>> content:
Thought: 我已經獲取了北京和廣州的天氣信息，現在可以回答用戶的問題了。


Final Answer: 北京的天氣情況為：溫度23.4°C，濕度43%，西南風，風速2.7米/秒，微風。廣州的天氣情況為：溫度24.2°C，濕度79%，東北風，風速1.3米/秒，微風。
>>> 最終答案: 北京的天氣情況為：溫度23.4°C，濕度43%，西南風，風速2.7米/秒，微風。廣州的天氣情況為：溫度24.2°C，濕度79%，東北風，風速1.3米/秒，微風。

總結

基于Function Calling和基于ReAct的工具調用有各自的優缺點：

1. Function Calling

無需設定系統提示詞，LLM根據tools定義即可觸發工具調用，token消耗較少
模型參數量相對較大。模型的訓練數據必須包含Function Calling相關的內容，以確保模型能夠理解和生成結構化輸出，結構化輸出更穩定
輸出結果較為容易處理
隱藏了推理過程，缺乏可解釋性

2. ReAct

需要設置復雜的系統提示詞，token消耗較多
對模型參數要求較低
輸出結果處理比Function Calling復雜
推理過程可見，更高的可解釋性

責任編輯：龐桂玉來源： AI大模型應用開發

成人免费xxxxx在线视频软件_久久精品久久久_亚洲国产精品久久久_天天色天天色_亚洲人成一区_欧美一级欧美三级在线观看

一文徹底搞懂智能體Agent基于ReAct的工具調用

前言

ReAct基礎

基于ReAct的工具調用流程和交互消息

1. 發起查詢請求

2. LLM返回Action獲取北京天氣

3. 處理函數調用獲取北京天氣

4. 把上下文信息以及函數調用結果發給LLM

5. LLM返回Action獲取廣州天氣

6. 處理函數調用獲取廣州天氣

7. 把上下文信息以及函數調用結果發給LLM

8. LLM生成最終回復

手搓Agent代碼實現基于ReAct的工具調用

1. 創建python環境

2. 設置API Key

3. 實現Agent代碼

總結

1. Function Calling

2. ReAct