使用uv结合crewai+火山引擎+deepseek v3 ，跑代码生成、RAG的例子

uv python install cpython-3.12.8-windows-x86_64-none

先用uv安装小于3.13版本的python

uv init crewai_demo3 --python 3.12.8
uv venv
.venv\Scripts\activate

llm = LLM(model="volcengine/ep-20250204220334-l2q5g", 
          api_key="xxxxxxxxxxxxxx")

然后火山引擎是这么搞

uv add crewai

接着添加依赖本身

因为我是需要跑代码

 https://docs.crewai.com/tools/codeinterpretertool 

uv add crewai[tools]

所以我又添加了tools的依赖

from crewai import LLM, Agent, Task, Crew, Process
from crewai_tools import CodeInterpreterTool

llm = LLM(model="volcengine/ep-20250204220334-l2q5g", 
          api_key="xxxxx")


# Initialize the tool
code_interpreter = CodeInterpreterTool()

# Define an agent that uses the tool
programmer_agent = Agent(
    role="Python Programmer",
    goal="Write and execute Python code to solve problems",
    backstory="An expert Python programmer who can write efficient code to solve complex problems.",
    tools=[code_interpreter],
    verbose=True,
    llm=llm
)

# Example task to generate and execute code
coding_task = Task(
    description="Write a Python function to calculate the Fibonacci sequence up to the 10th number and print the result.",
    expected_output="The Fibonacci sequence up to the 10th number.",
    agent=programmer_agent,
)

# Create and run the crew
crew = Crew(
    agents=[programmer_agent],
    tasks=[coding_task],
    verbose=True,
    process=Process.sequential,
)
result = crew.kickoff()

  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\sessions.py", line 637, in post
    return self.request("POST", url, data=data, json=json, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\sessions.py", line 589, in request
    resp = self.send(prep, **send_kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    return self._session.post(
           ^^^^^^^^^^^^^^^^^^^
  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\sessions.py", line 637, in post
    return self.request("POST", url, data=data, json=json, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\sessions.py", line 589, in request
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\sessions.py", line 589, in request
  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\sessions.py", line 589, in request
    resp = self.send(prep, **send_kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\sessions.py", line 703, in send
    r = adapter.send(request, **kwargs)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\development\crewai_demo3\.venv\Lib\site-packages\requests\adapters.py", line 698, in send
    raise SSLError(e, request=request)
requests.exceptions.SSLError: HTTPSConnectionPool(host='telemetry.crewai.com', port=4319): 
Max retries exceeded with url: /v1/traces (Caused by SSLError(SSLCertVerificationError(1, 
'[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: certificate has expired (_ssl.c:1000)')))
PS E:\development\crewai_demo3>

我特么的，报错啊。。。关键是，我没有要求你trace啊，太恶心了

说是访问telemetry.crewai.com遥测站点报错了。。。

无语

 https://docs.crewai.com/telemetry 

import os

# 设置环境变量（必须在导入OpenTelemetry库之前）
os.environ['OTEL_SDK_DISABLED'] = 'true'  # 注意值是字符串'true'

用这段代码把它关掉

# Initialize the tool
code_interpreter = CodeInterpreterTool(unsafe_mode=True)

还有就是记得让它可以在host上运行

from crewai import LLM, Agent, Task, Crew, Process
from crewai_tools import CodeInterpreterTool
import os

# 设置环境变量（必须在导入OpenTelemetry库之前）
os.environ['OTEL_SDK_DISABLED'] = 'true'  # 注意值是字符串'true'


llm = LLM(model="volcengine/ep-20250204220334-l2q5g", 
          api_key="xxxxxxxx")


# Initialize the tool
code_interpreter = CodeInterpreterTool(unsafe_mode=True)

# Define an agent that uses the tool
programmer_agent = Agent(
    role="Python Programmer",
    goal="Write and execute Python code to solve problems",
    backstory="An expert Python programmer who can write efficient code to solve complex problems.",
    tools=[code_interpreter],
    verbose=True,
    llm=llm
)

# Example task to generate and execute code
coding_task = Task(
    description="Write a Python function to calculate the Fibonacci sequence up to the 10th number and print the result.",
    expected_output="The Fibonacci sequence up to the 10th number.",
    agent=programmer_agent,
)

# Create and run the crew
crew = Crew(
    agents=[programmer_agent],
    tasks=[coding_task],
    verbose=True,
    process=Process.sequential,
)
result = crew.kickoff()

于是可以跑的通的代码如上所示

虽然确实输出结果了。。。但我有点害怕啊，这玩意真的是Tool Output么？

然后更有趣的是，crewai干脆做了一个插件市场

 https://app.composio.dev/apps 

 https://docs.crewai.com/tools/composiotool 

有269个工具可以直接调用

-----------------------

接着实验一下RAGTool，虽然它的RAGTool更像是一个开箱即用的东西，但确实看上去API非常简洁

 https://docs.crewai.com/tools/ragtool#agent-integration-example 

 https://docs.crewai.com/concepts/knowledge 

我们先从最简单的开始玩起吧

from crewai import Agent, Task, Crew, Process, LLM
from crewai.knowledge.source.string_knowledge_source import StringKnowledgeSource

import os

# 设置环境变量（必须在导入OpenTelemetry库之前）
os.environ['OTEL_SDK_DISABLED'] = 'true'  # 注意值是字符串'true'

#从环境变量取一下火山引擎的KEY出来
huoshan_key = os.getenv("HUOSHAN_API_KEY")

#litellm初始化火山引擎的方法
llm = LLM(model="volcengine/ep-20250204220334-l2q5g", 
          api_key=huoshan_key,temperature=0)


# Create a knowledge source
content = "Users name is John. He is 30 years old and lives in San Francisco."
string_source = StringKnowledgeSource(
    content=content,
)

# Create an agent with the knowledge store
agent = Agent(
    role="About User",
    goal="You know everything about the user.",
    backstory="""You are a master at understanding people and their preferences.""",
    verbose=True,
    allow_delegation=False,
    llm=llm,
    embedder={
        "provider": "ollama",
        "config": {
            "model": "bge-m3:latest",
            "api_key": "xxxx",
        }
    },
)
task = Task(
    description="Answer the following questions about the user: {question}",
    expected_output="An answer to the question.",
    agent=agent,
)

crew = Crew(
    agents=[agent],
    tasks=[task],
    verbose=True,
    process=Process.sequential,
    knowledge_sources=[string_source], # Enable knowledge by adding the sources here. You can also add more sources to the sources list.
)

result = crew.kickoff(inputs={"question": "What city does John live in and how old is he?"})

报错：Failed to upsert documents: APIStatusError.__init__() missing 2 required keyword-only arguments: 'response' and 'body'

然后看了一下是嵌入的问题：

 https://github.com/crewAIInc/crewAI/issues/2150 

from crewai import Agent, Task, Crew, Process, LLM
from crewai.knowledge.source.string_knowledge_source import StringKnowledgeSource

import os

# 设置环境变量（必须在导入OpenTelemetry库之前）
os.environ['OTEL_SDK_DISABLED'] = 'true'  # 注意值是字符串'true'

#从环境变量取一下火山引擎的KEY出来
huoshan_key = os.getenv("HUOSHAN_API_KEY")

#litellm初始化火山引擎的方法
llm = LLM(model="volcengine/ep-20250204220334-l2q5g", 
          api_key=huoshan_key,temperature=0)


# Create a knowledge source
content = "Users name is John. He is 30 years old and lives in San Francisco."
string_source = StringKnowledgeSource(
    content=content,
)

# Create an agent with the knowledge store
agent = Agent(
    role="About User",
    goal="You know everything about the user.",
    backstory="""You are a master at understanding people and their preferences.""",
    verbose=True,
    allow_delegation=False,
    llm=llm,
)
task = Task(
    description="Answer the following questions about the user: {question}",
    expected_output="An answer to the question.",
    agent=agent,
)

crew = Crew(
    agents=[agent],
    tasks=[task],
    verbose=True,
    process=Process.sequential,
    embedder={
        "provider": "ollama",
        "config": {
            "model": "bge-m3:latest",
            "api_key": "xxxx",
        }
    },
    knowledge_sources=[string_source], # Enable knowledge by adding the sources here. You can also add more sources to the sources list.
)

result = crew.kickoff(inputs={"question": "What city does John live in and how old is he?"})

修改代码成这样了，跑通了

要点其实也很简单，需要在crew的初始化那里，把kb的嵌入器配置好就行了

然后开始跑RAG_TOOLS这个，我才知道，它是集成了另外一个库

 https://docs.embedchain.ai/get-started/quickstart 

叫embedchain.ai，行吧，也算是工业外包了

from crewai import LLM, Agent
from crewai.project import agent
from crewai_tools import RagTool

import os

# 设置环境变量（必须在导入OpenTelemetry库之前）
os.environ['OTEL_SDK_DISABLED'] = 'true'  # 注意值是字符串'true'

#从环境变量取一下火山引擎的KEY出来
huoshan_key = os.getenv("HUOSHAN_API_KEY")

#litellm初始化火山引擎的方法
llm = LLM(model="volcengine/ep-20250204220334-l2q5g", 
          api_key=huoshan_key,temperature=0)


# Create a RAG tool with custom configuration
config = {
    "embedder": {
        "provider": "ollama",
        "config": {
            "model": "bge-m3:latest",
        }
    }
}

# Initialize the tool and add content
rag_tool = RagTool(llm=llm,config=config)
rag_tool.add(data_type="web_page", source="https://docs.crewai.com")

# Define an agent with the RagTool
@agent
def knowledge_expert(self) -> Agent:
    return Agent(
        config=self.agents_config["knowledge_expert"],
        allow_delegation=False,
        tools=[rag_tool],
    )

然后代码是有问题的，官方，估计是因为embedchain那边修改了

rag_tool.add(data_type="web_page", source="https://docs.crewai.com")

比如这么些

from crewai import LLM, Agent, Crew, Process, Task
from crewai.project import agent
from crewai_tools import RagTool

import os

# 设置环境变量（必须在导入OpenTelemetry库之前）
os.environ['OTEL_SDK_DISABLED'] = 'true'  # 注意值是字符串'true'

#从环境变量取一下火山引擎的KEY出来
huoshan_key = os.getenv("HUOSHAN_API_KEY")

#litellm初始化火山引擎的方法
llm = LLM(model="volcengine/ep-20250204220334-l2q5g", 
          api_key=huoshan_key,temperature=0)


# Create a RAG tool with custom configuration
config = {
    "embedder": {
        "provider": "ollama",
        "config": {
            "model": "bge-m3:latest",
        }
    }
}

# Initialize the tool and add content
rag_tool = RagTool(llm=llm,config=config)
rag_tool.add(data_type="web_page", source="https://docs.crewai.com")

# 定义 Project 类（关键修复点）
class MyProject():
    def __init__(self):
        super().__init__()
        # 定义 agents 配置（必须）
        self.agents_config = {
            "knowledge_expert": {
                "role": "知识专家",
                "goal": "准确回答关于 CrewAI 的问题",
                "backstory": "你是一个擅长使用 RAG 工具的技术文档专家",
                "verbose": True
            }
        }
    @agent  # 装饰器现在在 Project 类中
    def knowledge_expert(self) -> Agent:
        return Agent(
            config=self.agents_config["knowledge_expert"],
            allow_delegation=False,
            tools=[rag_tool],  # 使用全局 rag_tool
            llm=llm  # 显式传递 LLM
        )


# 初始化项目
project = MyProject()
knowledge_agent = project.knowledge_expert()  # 正确获取 Agent 实例

task = Task(
    description="Answer the following questions about the crewai: {question}",
    expected_output="An answer to the question.",
    agent=knowledge_agent,
)

crew = Crew(
    agents=[knowledge_agent],
    tasks=[task],
    verbose=True,
    process=Process.sequential,
)

result = crew.kickoff(inputs={"question": "用中文回答，crewai支持解析docx么？"})

笑死，跑是跑通了，但和没跑通也没多大区别

底层用的chroma，切分策略大到跟没有一样

哈哈哈哈

所以。。。你看吧

开箱即用的RAG，就不要期待太高。。。。没办法