1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

curl -X POST <BASE_URL>/chat/completions
-H "Authorization: Bearer <API_KEY>"
-H "Content-Type: application/json"
-d '{
"model": "<MODEL>",
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Hello!"}
],
"stream": false
}'

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

curl -X POST https://free.9e.nz/chat/completions
-H "Authorization: Bearer sk-cdf103d89a0628439c8889b546519d6bf0baedb700865e27e878a946e90760e9"
-H "Content-Type: application/json"
-d '{
"model": "opus4.6",
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Hello!"}
],
"stream": false
}'

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${DEEPSEEK_API_KEY}" \
  -d '{
        "model": "deepseek-chat",
        "messages": [
          {"role": "system", "content": "You are a helpful assistant."},
          {"role": "user", "content": "Hello!"}
        ],
        "stream": false
      }'

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-xxxxxxxxxxxxxxxxxxxxxxx" \
  -d '{
        "model": "deepseek-chat",
        "messages": [
          {"role": "system", "content": "You are a helpful assistant."},
          {"role": "user", "content": "Hello!"}
        ],
        "stream": false
      }'

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \
-H "Authorization: Bearer $DASHSCOPE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
    "model": "qwen-plus",
    "messages": [
        {
            "role": "system",
            "content": "You are a helpful assistant."
        },
        {
            "role": "user",
            "content": "你是谁？"
        }
    ]
}'

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \
-H "Authorization: Bearer sk-xxxxxxxxxxxxxxxxx" \
-H "Content-Type: application/json" \
-d '{
    "model": "qwen-plus",
    "messages": [
        {
            "role": "system",
            "content": "You are a helpful assistant."
        },
        {
            "role": "user",
            "content": "你是谁？"
        }
    ]
}'

1

OllamaSetup.exe /DIR=你要安装的目录位置

1

OLLAMA_MODELS=你想要保存模型的目录

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

  ollama serve      # Start ollama
  ollama create     # Create a model from a Modelfile
  ollama show       # Show information for a model
  ollama run        # Run a model
  ollama stop       # Stop a running model
  ollama pull       # Pull a model from a registry
  ollama push       # Push a model to a registry
  ollama list       # List models
  ollama ps         # List running models
  ollama cp         # Copy a model
  ollama rm         # Remove a model
  ollama help       # Help about any command

1
2
3
4
5

curl http://localhost:11434/api/chat \
  -d '{
    "model": "glm-5.1:cloud",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

1
2
3
4
5

 curl http://localhost:11434/api/chat -d '{
    "model": "deepseek-r1:8b",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": false
  }'

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18

POST /chat/completions HTTP/1.1
Content-Type: application/json
Authorization: Bearer sk-xxxxxxxxxxxxxxxxxxxxxxxx
User-Agent: Mozilla/5.0 (SymbianOS/9.1; U; de) AppleWebKit/413 (KHTML, like Gecko) Safari/413
Accept: */*
Host: api.deepseek.com
Content-Length: 822

{
  "model": "deepseek-chat",
  "messages": [
    { "role": "system", "content": "You are a helpful assistant." },
    { "role": "user", "content": "Hello!你是谁？我是heima" },
    { "role": "assistant", "content": "你好，heima！我是DeepSeek，由深度求索公司创造的AI助手。很高兴认识你！😊\n\n我是一个纯文本模型，可以帮你解答各种问题、协助处理文档、进行对话交流等等。虽然我不支持多模态识别，但我可以处理你上传的文件（图像、txt、pdf、ppt、word、excel等），从中读取文字信息来帮助你。\n\n有什么我可以帮你的吗？无论是学习、工作还是日常问题，我都很乐意为你提供帮助！"},
    { "role": "user", "content": "你记得我是谁吗" }
  ],
  "stream": false
}

1

setx UV_DEFAULT_INDEX "https://pypi.tuna.tsinghua.edu.cn/simple"

1

https://mirrors.aliyun.com/pypi/simple/

1

https://mirrors.cloud.tencent.com/pypi/simple/

1

https://mirrors.volces.com/pypi/simple/

1

https://mirrors.huaweicloud.com/repository/pypi/simple/

1

https://pypi.tuna.tsinghua.edu.cn/simple/

1

https://pypi.mirrors.ustc.edu.cn/simple/

1

uv add notebook

1
2

from langchain.chat_models import init_chat_model
model = init_chat_model(model="deepseek-chat")

1
2
3
4
5
6
7
8
9

system_prompt = """
 你是一名私人厨师。收到用户提供的食材照片或清单后，请按以下流程操作：
 1.识别和评估食材：若用户提供照片，首先辨识所有可见食材。基于食材的外观状态，评估其新鲜度与可用量，整理出一份“当前可用食材清单”。
 2.智能食谱检索：优先调用 web_search 工具，以“可用食材清单”为核心关键词，查找可行菜谱。
 3.多维度评估与排序：从营养价值和制作难度两个维度对检索到的候选食谱进行量化打分，并根据得分排序，制作简单且营养丰富的排名靠前。
 4.结构化方案输出：把排序后的食谱整理为一份结构清晰的建议报告，要包含食谱信息、得分、推荐理由、食谱的参考图片，帮助用户快速做出决策。
 
 请严格按照流程，优先调用 web_search 工具搜索食谱，搜索不到的情况下才能自己发挥。
 """

1
2
3
4
5
6
7
8

# 阿里百炼
 DASHSCOPE_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
 DASHSCOPE_API_KEY=sk-36b7386d28524af486314f6156842c0b
 
 
 # tools
 ## Tavily工具
 TAVILY_API_KEY=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

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

[project]
 name = "food-recipe-recommender"
 version = "0.1.0"
 description = "AI-powered recipe recommender based on uploaded food images"
 readme = "README.md"
 requires-python = ">=3.10"
 dependencies = [
     "langchain>=0.3.0",
     "langchain-community>=0.3.0",
     "langchain-openai>=0.2.0",
     "langchain-core>=0.3.0",
     "streamlit>=1.40.0",
     "pillow>=11.0.0",
     "python-dotenv>=1.0.0",
     "tavily-python>=0.5.0",
     "langchain-tavily>=0.1.0",
     "fastapi>=0.109.0",
     "uvicorn>=0.27.0",
     "python-multipart>=0.0.6",
     "alibabacloud-oss-v2>=1.2.4",
     "langgraph-checkpoint-sqlite>=3.0.3",
 ]
 
 [tool.uv]
 dev-dependencies = [
     "pytest>=8.0.0",
     "black>=24.0.0",
 ]

1
2
3

# langsmith
 LANGSMITH_API_KEY=
 LANGSMITH_TRACING=true

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

from langchain.chat_models import init_chat_model
 from langchain_tavily import TavilySearch
 from langchain.agents import create_agent
 import os
 
 # 1.加载环境变量
 from dotenv import load_dotenv
 load_dotenv()
 
 # 2.web搜索工具，使用tavily作为web搜索工具
 web_search = TavilySearch(
     max_results=5,
     topic="general"
 )
 
 # 3.多模态模型
 model = init_chat_model(
     model="qwen3.5-plus",  # 模型名称，这里选择qwen3.5-plus，这是一个多模态模型，支持图片、文本、音频、视频
     model_provider="openai",
     base_url=os.getenv("DASHSCOPE_BASE_URL"),
     api_key=os.getenv("DASHSCOPE_API_KEY")
 )
 
 # 4.Agent系统提示词
 system_prompt = """
 你是一名私人厨师。收到用户提供的食材照片或清单后，请按以下流程操作：
 1.识别和评估食材：若用户提供照片，首先辨识所有可见食材。基于食材的外观状态，评估其新鲜度与可用量，整理出一份“当前可用食材清单”。
 2.智能食谱检索：优先调用 web_search 工具，以“可用食材清单”为核心关键词，查找可行菜谱。
 3.多维度评估与排序：从营养价值和制作难度两个维度对检索到的候选食谱进行量化打分，并根据得分排序，制作简单且营养丰富的排名靠前。
 4.结构化方案输出：把排序后的食谱整理为一份结构清晰的建议报告，要包含食谱信息、得分、推荐理由、食谱的参考图片，帮助用户快速做出决策。
 
 请严格按照流程，优先调用 web_search 工具搜索食谱，搜索不到的情况下才能自己发挥。
 """
 
 # 5.创建Agent
 agent = create_agent(
     model=model,  # 模型
     tools=[web_search],  # 工具
     system_prompt=system_prompt  # 系统提示词
 )

1
2
3
4
5
6
7

{
     "dependencies": ["."],
     "graphs": {
         "chief_agent": "./app/agents/personal_chief.py:agent"
     },
     "env": ".env"
 }

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

app/
   ├── main.py                    # FastAPI 入口，配置路由和静态文件
   │
   ├── agents/
   │   └── personal_chief.py      # AI 代理核心逻辑
   │ 
   ├── api/
   │   └── v1/
   │       ├── chat.py             # 对话 API
   │       │   ├── POST /chat/stream     流式对话
   │       │   ├── GET  /chat/messages   获取历史
   │       │   └── DELETE /chat/messages 清空历史
   │       │
   │       └── oss.py              # OSS 上传签名 URL
   │
   ├── models/
   │   └── schemas.py              # Pydantic 数据模型，请求/响应数据结构定义  
   │
   ├── common/
   │   └── logger.py               # 日志配置
   │
   └── static/                     # Next.js 编译产出的静态网页 
       ├── index.html              # 前端入口
       ├── _next/                  # Next.js 构建资源
       └── ...                     # 其他静态资源

1
2
3
4

# 阿里OSS
 OSS_ACCESS_KEY_ID=xxxxxxxxxxxx
 OSS_ACCESS_KEY_SECRET=xxxxxxxxxxxx
 OSS_BUCKET=persxxxxxxxxxxx

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51

from langchain.chat_models import init_chat_model
from langchain_tavily import TavilySearch
from langchain.agents import create_agent
import os
from langgraph.checkpoint.sqlite import SqliteSaver
import sqlite3

# 1.加载环境变量
from dotenv import load_dotenv
load_dotenv()

# 2.web搜索工具，使用tavily作为web搜索工具
web_search = TavilySearch(
    max_results=5,
    topic="general"
)

# 3.多模态模型
model = init_chat_model(
    model="qwen3.5-plus",  # 模型名称，这里选择qwen3.5-plus，这是一个多模态模型，支持图片、文本、音频、视频
    model_provider="openai",
    base_url=os.getenv("DASHSCOPE_BASE_URL"),
    api_key=os.getenv("DASHSCOPE_API_KEY")
)

# 4.初始化checkpointer
# 连接sqlite
connection = sqlite3.connect("../db/personal_chief.db", check_same_thread=False)
# 初始化checkpointer
checkpointer = SqliteSaver(connection)
# 自动建表
checkpointer.setup()

# 5.Agent系统提示词
system_prompt = """
 你是一名私人厨师。收到用户提供的食材照片或清单后，请按以下流程操作：
 1.识别和评估食材：若用户提供照片，首先辨识所有可见食材。基于食材的外观状态，评估其新鲜度与可用量，整理出一份“当前可用食材清单”。
 2.智能食谱检索：优先调用 web_search 工具，以“可用食材清单”为核心关键词，查找可行菜谱。
 3.多维度评估与排序：从营养价值和制作难度两个维度对检索到的候选食谱进行量化打分，并根据得分排序，制作简单且营养丰富的排名靠前。
 4.结构化方案输出：把排序后的食谱整理为一份结构清晰的建议报告，要包含食谱信息、得分、推荐理由、食谱的参考图片，帮助用户快速做出决策。
 
 请严格按照流程，优先调用 web_search 工具搜索食谱，搜索不到的情况下才能自己发挥。
 """

# 6.创建Agent
agent = create_agent(
    model=model,  # 模型
    tools=[web_search],  # 工具
    checkpointer=checkpointer,  # 记忆
    system_prompt=system_prompt  # 系统提示词
)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19

from fastapi import APIRouter

router = APIRouter()

@router.post("/chat/stream")
async def chat_endpoint(request: ChatRequest):
    """流式聊天接口"""
    pass

@router.get("/chat/messages")
async def get_chat_messages(thread_id: str):
    """获取历史消息"""
    pass


@router.delete("/chat/messages")
async def clear_chat_messages(thread_id: str):
    """清空历史消息"""
    pass

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121

from langchain.chat_models import init_chat_model
from langchain_core.messages import HumanMessage, AIMessageChunk, AIMessage
from langchain_core.tools import tool
from langchain_tavily import TavilySearch
from langchain.agents import create_agent
from app.common.logger import logger
import os
from langgraph.checkpoint.sqlite import SqliteSaver
import sqlite3

# 加载环境变量
from dotenv import load_dotenv

load_dotenv()

# web搜索工具，使用tavily作为web搜索工具
tavily = TavilySearch(
    max_results=5,
    topic="general"
)


# 多模态模型
model = init_chat_model(
    model="qwen3-omni-flash",  # 模型名称，这里选择qwen3.5-plus，这是一个多模态模型，支持图片、文本、音频、视频
    model_provider="openai",
    base_url=os.getenv("DASHSCOPE_BASE_URL"),
    api_key=os.getenv("DASHSCOPE_API_KEY")
)


# 初始化checkpointer
checkpointer = SqliteSaver(sqlite3.connect("db/personal_chief.db", check_same_thread=False))
# 自动建表
checkpointer.setup()

# Agent系统提示词
system_prompt = """
 你是一名私人厨师。收到用户提供的食材照片或清单后，请按以下流程操作：
 1.识别和评估食材：若用户提供照片，首先辨识所有可见食材。基于食材的外观状态，评估其新鲜度与可用量，整理出一份“当前可用食材清单”。
 2.智能食谱检索：优先调用 web_search 工具，以“可用食材清单”为核心关键词，查找可行菜谱。
 3.多维度评估与排序：从营养价值和制作难度两个维度对检索到的候选食谱进行量化打分，并根据得分排序，制作简单且营养丰富的排名靠前。
 4.结构化方案输出：把排序后的食谱整理为一份结构清晰的建议报告，要包含食谱信息、得分、推荐理由、食谱的参考图片，帮助用户快速做出决策。
 
 请严格按照流程，优先调用 web_search 工具搜索食谱，搜索不到的情况下才能自己发挥。
 """

# 创建代理
agent = create_agent(
    model=model,  # 模型
    tools=[tavily],  # 工具
    checkpointer=checkpointer,  # 记忆
    system_prompt=system_prompt  # 系统提示词
)

# 流式对话
async def search_recipes(prompt: str, image: str, thread_id: str):
    """调用agent搜索食谱"""
    logger.info(f"[用户]: {prompt}, image: {image}, thread_id: {thread_id}")
    try:
        # 判断是否有图片，封装不同格式的消息
        if not image or image.strip() == "":
            message = HumanMessage(content=prompt)
        else:
            message = HumanMessage(content=[
                {"type": "image", "url": image},
                {"type": "text", "text": prompt}
            ])

        # 流式调用Agent
        for chunk, metadata in agent.stream(
            {"messages": [message]},
            {"configurable": {"thread_id": thread_id}},
            stream_mode="messages"
        ):
            if isinstance(chunk, AIMessageChunk) and chunk.content:
                yield chunk.content

    except Exception as e:
        logger.error(f"\n[错误]: {str(e)}")
        yield "信息检索失败，试试看手动输入食物列表？"

# 清空会话
def clear_messages(thread_id: str):
    """清空会话"""
    logger.info(f"清空历史消息，thread_id: {thread_id}")
    checkpointer.delete_thread(thread_id)

# 查询会话历史
def get_messages(thread_id: str) -> list[dict[str, str]]:
    """获取会话历史"""
    logger.info(f"获取历史消息，thread_id: {thread_id}")

    # 根据 thread_id 查询 checkpoint
    checkpoint = checkpointer.get({"configurable": {"thread_id": thread_id}})

    # 如果不存在，返回空列表
    if not checkpoint:
        return []

    # 安全获取 messages
    channel_values = checkpoint.get("channel_values")
    if not channel_values:
        return []

    messages = channel_values.get("messages", [])
     if not messages:
         return []
 
     # 转换消息格式
     result = []
     for msg in messages:
         if not msg.content:
             continue
 
         if isinstance(msg, HumanMessage):
             result.append({"role": "user", "content": msg.content})
         elif isinstance(msg, AIMessage):
             result.append({"role": "assistant", "content": msg.content})
 
     return result

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

from fastapi import APIRouter
from app.models.schemas import ChatRequest
from fastapi.responses import StreamingResponse
from app.agents.personal_chief import search_recipes, get_messages, clear_messages


router = APIRouter()


@router.post("/chat/stream")
async def chat_endpoint(request: ChatRequest):
    """流式对话"""
    return StreamingResponse(
        search_recipes(request.message, request.image_url, request.thread_id),
        media_type="text/event-stream"
    )


@router.get("/chat/messages")
async def get_chat_messages(thread_id: str):
    """获取历史消息"""
    messages = get_messages(thread_id)
    return {"messages": messages}


@router.delete("/chat/messages")
async def clear_chat_messages(thread_id: str):
    """清空历史消息"""
    clear_messages(thread_id)
    return {"success": True}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

# 阿里百炼
DASHSCOPE_BASE_URL=
DASHSCOPE_API_KEY=


# tools
## Tavily工具
TAVILY_API_KEY=

# LangSmith
LANGSMITH_API_KEY=
LANGSMITH_TRACING=true

# project name
LANGSMITH_PROJECT=

# 阿里OSS
OSS_ACCESS_KEY_ID=
OSS_ACCESS_KEY_SECRET=3cZXcWqoP5C1jmDrDiuDT5MX
OSS_BUCKET=

1

docker build --network=host -t food-recipe-recommender:1.0.0 .

1
2
3
4
5
6
7

docker run -d \
  --name food-recipe-recommender \
  --restart unless-stopped \
  -p 8001:8001 \
  --env-file .env \
  -v /opt/food-recipe/db:/app/db \
  food-recipe-recommender:1.0.0

1
2
3

docker login --username=xvsf539 crpi-z2yvaep8ppb79obm.cn-hangzhou.personal.cr.aliyuncs.com
docker tag food-recipe-recommender:1.0.0 crpi-z2yvaep8ppb79obm.cn-hangzhou.personal.cr.aliyuncs.com/docker_fast_test/docker-a:1.0.0
docker push crpi-z2yvaep8ppb79obm.cn-hangzhou.personal.cr.aliyuncs.com/docker_fast_test/docker-a:1.0.0

1
2
3
4
5
6
7
8
9

docker pull crpi-z2yvaep8ppb79obm.cn-hangzhou.personal.cr.aliyuncs.com/docker_fast_test/docker-a:1.0.0

docker run -d \
  --name food-recipe-recommender \
  --restart unless-stopped \
  -p 8001:8001 \
  --env-file /opt/food-recipe/.env \
  -v /opt/food-recipe/db:/app/db \
  crpi-z2yvaep8ppb79obm.cn-hangzhou.personal.cr.aliyuncs.com/docker_fast_test/docker-a:1.0.0