Agno - 轻量级 Python 多智能体系统框架

# 创建虚拟环境python3 -m venv .venvsource .venv/bin/activate
# 安装核心依赖pip install agno openai duckduckgo-search

# 向量数据库支持pip install pgvector lancedb qdrant-client
# 多模态扩展pip install elevenlabs opencv-python

# 安装开发依赖pip install -U agno[dev]
# 运行Playgroundpython playground.py

from agno.agent import Agentfrom agno.models.openai import OpenAIChat
agent = Agent(    model=OpenAIChat(id="gpt-4"),    instructions="你是一个热情的新闻记者")agent.print_response("分享一则纽约新闻")

from agno.agent import Agentfrom agno.knowledge.pdf import PDFKnowledgeBase
knowledge = PDFKnowledgeBase(path="data/docs")agent = Agent(knowledge=knowledge)agent.print_response("文档中提到哪些关键技术?")

from agno.agent import Agentfrom agno.tools.yfinance import YFinanceTools
agent = Agent(tools=[YFinanceTools()])agent.print_response("苹果公司当前股价是多少?")

class Agent:    def __init__(        self,        model: BaseModel,          # 语言模型         tools: List[BaseTool] = [], # 工具集        knowledge: BaseKnowledge = None,  # 知识库        storage: BaseStorage = None,      # 会话存储        reasoning: bool = False    # 启用推理    ):        self.components = {            'model': model,            'tools': ToolManager(tools),            'knowledge': knowledge,            'storage': storage        }

def search_knowledge(query):    # 混合检索流程    vector_results = vector_db.semantic_search(query)    keyword_results = fulltext_search(query)        # 结果重排序    combined = hybrid_reranker(vector_results, keyword_results)        # 上下文注入    return format_for_prompt(combined[:5])

async def execute_tool(tool_name, args):    tool = self.tools.get(tool_name)        # 异步执行    result = await tool.execute(args)        # 结果验证    if not validate_result(result):        raise ToolExecutionError        return format_tool_output(result)