在前面的文章裡《一文探秘LLM應用開發(26)-Prompt(架構模式之Agent框架AutoGPT、AutoGen等)》，筆者曾介紹了一些諸如AutoGen、ChatDev這樣的多Agent框架。最近行業內又出現一款不錯的框架——CrewAI，可謂是站在AutoGen這樣的框架肩膀上，以實際投產為目標，兼具AutoGen對話代理的靈活性和ChatDev的領域流程化的優點，規避了AutoGen缺乏框架層面的流程支持和ChatDev領域流程過於狹窄不夠泛化的問題，支持動態的泛場景的流程設計，能夠無縫適應開發和生產工作流程。也就是說，在CrewAI中可以像AutoGen一樣結合場景定義自己的角色，又能像ChatDev那樣約定一定的流程執行，讓這些Agents能夠更好完成特定的復雜目標。該項目現已收獲3.9K星，並獲得ProductHunt排名第二的好位置。
https://github.com/joaomdmoura/crewAI
CrewAI具備以下一些關鍵feature：
- 基於角色的Agent設計：定制具有特定角色、目標和工具的Agent。
- Agent間自主委托：Agent可自主委派任務，並在Agent之間進行詢問，從而提高解決問題的效率。
- 靈活的任務管理：使用可定制的工具定義任務，並將其動態分配給Agent。
- 流程驅動（最大的亮點）：目前僅支持順序（sequential）任務執行，但在規劃中的有更高階的流程定義，如共識和分層流程。
創建一個流程大體如下：
import os
from crewai import Agent, Task, Crew, Process
os.environ["OPENAI_API_KEY"] = "YOUR KEY"
# You can choose to use a local model through Ollama for example.
#
# from langchain.llms import Ollama
# ollama_llm = Ollama(model="openhermes")
# Install duckduckgo-search for this example:
# !pip install -U duckduckgo-search
from langchain.tools import DuckDuckGoSearchRun
search_tool = DuckDuckGoSearchRun()
# Define your agents with roles and goals
researcher = Agent(
role='Senior Research Analyst',
goal='Uncover cutting-edge developments in AI and data science',
backstory="""You work at a leading tech think tank.
Your expertise lies in identifying emerging trends.
You have a knack for dissecting complex data and presenting
actionable insights.""",
verbose=True,
allow_delegation=False,
tools=[search_tool]
# You can pass an optional llm attribute specifying what mode you wanna use.
# It can be a local model through Ollama / LM Studio or a remote
# model like OpenAI, Mistral, Antrophic of others (https://python.langchain.com/docs/integrations/llms/)
#
# Examples:
# llm=ollama_llm # was defined above in the file
# llm=ChatOpenAI(model_name="gpt-3.5", temperature=0.7)
)
writer = Agent(
role='Tech Content Strategist',
goal='Craft compelling content on tech advancements',
backstory="""You are a renowned Content Strategist, known for
your insightful and engaging articles.
You transform complex concepts into compelling narratives.""",
verbose=True,
allow_delegation=True,
# (optional) llm=ollama_llm
)
# Create tasks for your agents
task1 = Task(
description="""Conduct a comprehensive analysis of the latest advancements in AI in 2024.
Identify key trends, breakthrough technologies, and potential industry impacts.
Your final answer MUST be a full analysis report""",
agent=researcher
)
task2 = Task(
description="""Using the insights provided, develop an engaging blog
post that highlights the most significant AI advancements.
Your post should be informative yet accessible, catering to a tech-savvy audience.
Make it sound cool, avoid complex words so it doesn't sound like AI.
Your final answer MUST be the full blog post of at least 4 paragraphs.""",
agent=writer
)
# Instantiate your crew with a sequential process
crew = Crew(
agents=[researcher, writer],
tasks=[task1, task2],
verbose=2,# You can set it to 1 or 2 to different logging levels
process=Process.sequential
)
# Get your crew to work!
result = crew.kickoff()
print("######################")
print(result)如上，定義了兩個Agent，researcher負責調研收集信息，writer基於這些信息寫作，然後根據任務順序執行的流程進行執行。對於任務本身也可以進行自定義，比如指定特定的信源收集信息，如下是在reddit的LocalLLaMA版上獲取信息。
# pip install praw
from langchain.tools import tool
class BrowserTool():
@tool("Scrape reddit content")
def scrape_reddit(max_comments_per_post=5):
"""Useful to scrape a reddit content"""
reddit = praw.Reddit(
client_id="your-client-id",
client_secret="your-client-secret",
user_agent="your-user-agent",
)
subreddit = reddit.subreddit("LocalLLaMA")
scraped_data = []
for post in subreddit.hot(limit=10):
post_data = {"title": post.title, "url": post.url, "comments": []}
try:
post.comments.replace_more(limit=0) # Load top-level comments only
comments = post.comments.list()
if max_comments_per_post is not None:
comments = comments[:5]
for comment in comments:
post_data["comments"].append(comment.body)
scraped_data.append(post_data)
except praw.exceptions.APIException as e:
print(f"API Exception: {e}")
time.sleep(60) # Sleep for 1 minute before retrying
return scraped_data而使用它隻需要將search_tool更換為BrowserTool().scrape_reddit。同時，它也支持將human作為工具編織在Agent中。對於這樣做的好處可以參考文章《人充當LLM Agent的工具（Human-In-The-Loop ），提升復雜問題解決成功率》。
# Define your agents with roles and goals
researcher = Agent(
role='Senior Research Analyst',
goal='Uncover cutting-edge developments in AI and data science in',
backstory="""You are a Senior Research Analyst at a leading tech think tank.
Your expertise lies in identifying emerging trends and technologies in AI and
data science. You have a knack for dissecting complex data and presenting
actionable insights.""",
verbose=True,
allow_delegation=False,
# Passing human tools to the agent
tools=[search_tool] human_tools
)另外，為了節省token費用以及隱私性，還可以對接本地LLM平臺Ollama，有關Ollama介紹參看《一文探秘LLM應用開發(17)-模型部署與推理(框架工具-ggml、mlc-llm、ollama)》，安裝，配置完成後可通過以下方式集成，Ollama上有很多本地模型，推薦使用當紅的mistral模型。
from langchain.llms import Ollama
ollama_openhermes = Ollama(model="openhermes")
# Pass Ollama Model to Agents: When creating your agents within the CrewAI framework, you can pass the Ollama model as an argument to the Agent constructor. For instance:
local_expert = Agent(
role='Local Expert at this city',
goal='Provide the BEST insights about the selected city',
backstory="""A knowledgeable local guide with extensive information
about the city, it's attractions and customs""",
tools=[
SearchTools.search_internet,
BrowserTools.scrape_and_summarize_website,
],
llm=ollama_openhermes, # Ollama model passed here
verbose=True
)更多官方案例：
https://github.com/joaomdmoura/crewAI-examples