# macOS / Linux（一行搞定）
curl -fsSL https://ollama.com/install.sh | sh
# Windows：前往 https://ollama.com/download 下載安裝包

ollama --version
# 輸出：ollama version is 0.6.x

# 推薦：Qwen3 8B（中文能力最強的小模型之一）
ollama pull qwen3:8b
# 備選：Llama 4 Scout 17B（Meta 最新開源模型）
ollama pull llama4:scout17b
# 輕量選擇：Qwen3 4B（4GB 內(nèi)存就能跑）
ollama pull qwen3:4b
# 查看已下載的模型
ollama list

pip install ollama

import ollama
# 發(fā)送一條消息，獲取回復
response = ollama.chat(
    model="qwen3:8b",
    messages=[
        {"role": "user", "content": "用 Python 寫一個快速排序算法"}
    ]
)
print(response["message"]["content"])

import ollama
# stream=True 開啟流式輸出，體驗更絲滑
stream = ollama.chat(
    model="qwen3:8b",
    messages=[
        {"role": "user", "content": "解釋一下 Python 的裝飾器"}
    ],
    stream=True
)
for chunk in stream:
    print(chunk["message"]["content"], end="", flush=True)

import ollama
# 維護對話歷史
messages = []
def chat(user_input: str) -> str:
    messages.append({"role": "user", "content": user_input})
    response = ollama.chat(model="qwen3:8b", messages=messages)
    assistant_msg = response["message"]["content"]
    messages.append({"role": "assistant", "content": assistant_msg})
    return assistant_msg
# 多輪對話
print(chat("你好，我正在學 Python"))
print("---")
print(chat("剛才我說我在學什么來著？"))  # 它能記住上下文

local-ai-assistant/
├── assistant.py      # 主程序
├── config.py         # 配置文件
└── chat_history.json # 對話歷史（自動生成）

# config.py
MODEL_NAME = "qwen3:8b"          # 使用的模型
SYSTEM_PROMPT = """你是一個專業(yè)的 AI 編程助手。
你擅長 Python、數(shù)據(jù)分析、自動化辦公。
回答要簡潔實用，代碼要可以直接運行。"""
MAX_HISTORY = 20                   # 最多保留的對話輪數(shù)

# assistant.py
import json
import ollama
from datetime import datetime
from config import MODEL_NAME, SYSTEM_PROMPT, MAX_HISTORY
def load_history(filepath="chat_history.json"):
    """加載對話歷史"""
    try:
        with open(filepath, "r", encoding="utf-8") as f:
            return json.load(f)
    except FileNotFoundError:
        return []
def save_history(messages, filepath="chat_history.json"):
    """保存對話歷史"""
    with open(filepath, "w", encoding="utf-8") as f:
        json.dump(messages, f, ensure_ascii=False, indent=2)
def chat_stream(user_input: str, messages: list) -> str:
    """流式對話"""
    messages.append({"role": "user", "content": user_input})
    print(f"\n\033[92m?? AI:\033[0m ", end="", flush=True)
    full_response = ""
    stream = ollama.chat(model=MODEL_NAME, messages=messages, stream=True)
    for chunk in stream:
        content = chunk["message"]["content"]
        print(content, end="", flush=True)
        full_response += content
    print("\n")
    messages.append({"role": "assistant", "content": full_response})
    # 限制歷史長度
    if len(messages) > MAX_HISTORY * 2 + 1:
        messages = [messages[0]] + messages[-(MAX_HISTORY * 2):]
    return full_response
def main():
    print("=" * 50)
    print(f"  本地 AI 助手  |  模型: {MODEL_NAME}")
    print("  輸入 /clear 清空歷史 | /quit 退出")
    print("=" * 50)
    # 初始化對話
    messages = load_history()
    if not messages:
        messages = [{"role": "system", "content": SYSTEM_PROMPT}]
    while True:
        user_input = input("\n\033[94m?? 你:\033[0m ").strip()
        if not user_input:
            continue
        if user_input == "/quit":
            save_history(messages)
            print("?? 再見！對話已保存。")
            break
        if user_input == "/clear":
            messages = [{"role": "system", "content": SYSTEM_PROMPT}]
            save_history(messages)
            print("???  對話歷史已清空。")
            continue
        chat_stream(user_input, messages)
        save_history(messages)
if __name__ == "__main__":
    main()

python assistant.py

import ollama
import json
# 定義工具函數(shù)
def get_weather(city: str) -> str:
    """模擬獲取天氣（可替換為真實 API）"""
    weather_data = {"北京": "晴天 22°C", "上海": "多云 18°C", "深圳": "大雨 28°C"}
    return weather_data.get(city, "未查詢到該城市天氣")
def calculate(expression: str) -> str:
    """安全計算數(shù)學表達式"""
    try:
        allowed = set("0123456789+-*/.() ")
        if all(c in allowed for c in expression):
            return str(eval(expression))
        return "不安全的表達式"
    except Exception as e:
        return f"計算錯誤: {e}"
# 注冊工具
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "獲取指定城市的天氣信息",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {"type": "string", "description": "城市名稱"}
                },
                "required": ["city"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "calculate",
            "description": "計算數(shù)學表達式",
            "parameters": {
                "type": "object",
                "properties": {
                    "expression": {"type": "string", "description": "數(shù)學表達式"}
                },
                "required": ["expression"]
            }
        }
    }
]
# 工具映射
tool_map = {
    "get_weather": get_weather,
    "calculate": calculate
}
# 對話
messages = [{"role": "user", "content": "北京天氣怎么樣？順便幫我算一下 25*36+128"}]
response = ollama.chat(model="qwen3:8b", messages=messages, tools=tools)
# 處理工具調(diào)用
if response["message"].get("tool_calls"):
    for tool_call in response["message"]["tool_calls"]:
        func_name = tool_call["function"]["name"]
        func_args = tool_call["function"]["arguments"]
        print(f"?? 調(diào)用工具: {func_name}({func_args})")
        result = tool_map[func_name](**func_args)
        print(f"   結(jié)果: {result}")

# 如果你已有使用 OpenAI SDK 的代碼，只需改一行
from openai import OpenAI
client = OpenAI(
    base_url="http://localhost:11434/v1",  # 指向 Ollama
    api_key="ollama"                        # 隨意填，不影響
)
response = client.chat.completions.create(
    model="qwen3:8b",
    messages=[
        {"role": "user", "content": "Hello, who are you?"}
    ]
)
print(response.choices[0].message.content)

Ollama 本地 AI 的核心優(yōu)勢：
├── 零成本 —— 不花一分錢 API 費用
├── 零風險 —— 數(shù)據(jù)永遠不會離開你的電腦
├── 零門檻 —— pip install + ollama pull，5 分鐘上手
└── 零妥協(xié) —— 兼容 OpenAI API，現(xiàn)有生態(tài)無縫切換