Skip to main content
使用 Connery 工具包和工具,您可以将 Connery Actions 集成到您的 LangChain 智能体中。

什么是 Connery?

Connery 是一个面向 AI 的开源插件基础设施。 通过 Connery,您可以轻松创建包含一组操作的自定义插件,并将其无缝集成到您的 LangChain 智能体中。 Connery 将处理关键方面,例如运行时、授权、密钥管理、访问管理、审计日志和其他重要功能。 此外,Connery 由社区支持,提供多样化的现成开源插件以方便使用。 了解更多关于 Connery 的信息:

设置

安装

您需要安装 langchain_community 包才能使用 Connery 工具。
pip install -qU langchain-community

凭证

要在您的 LangChain 智能体中使用 Connery Actions,您需要做一些准备:
  1. 使用 快速开始 指南设置 Connery runner。
  2. 安装您想要在智能体中使用的所有包含操作的插件。
  3. 设置环境变量 CONNERY_RUNNER_URLCONNERY_RUNNER_API_KEY,以便工具包能够与 Connery Runner 通信。
import getpass
import os

for key in ["CONNERY_RUNNER_URL", "CONNERY_RUNNER_API_KEY"]:
    if key not in os.environ:
        os.environ[key] = getpass.getpass(f"Please enter the value for {key}: ")

工具包

在下面的示例中,我们创建了一个使用两个 Connery Actions 来总结公共网页并通过电子邮件发送摘要的智能体:
  1. 来自 摘要 插件的 总结公共网页 操作。
  2. 来自 Gmail 插件的 发送电子邮件 操作。
您可以在这里看到此示例的 LangSmith 追踪。
import os

from langchain.agents import create_agent
from langchain_community.agent_toolkits.connery import ConneryToolkit
from langchain_community.tools.connery import ConneryService
from langchain_openai import ChatOpenAI

# Specify your Connery Runner credentials.
os.environ["CONNERY_RUNNER_URL"] = ""
os.environ["CONNERY_RUNNER_API_KEY"] = ""

# Specify OpenAI API key.
os.environ["OPENAI_API_KEY"] = ""

# Specify your email address to receive the email with the summary from example below.
recepient_email = "test@example.com"

# Create a Connery Toolkit with all the available actions from the Connery Runner.
connery_service = ConneryService()
connery_toolkit = ConneryToolkit.create_instance(connery_service)

# Use OpenAI Functions agent to execute the prompt using actions from the Connery Toolkit.
llm = ChatOpenAI(temperature=0)

agent = create_agent(
    model=llm,
    tools=connery_toolkit.get_tools(),
    verbose=True,
)
result = agent.invoke({
    "input": f"""Make a short summary of the webpage http://www.paulgraham.com/vb.html in three sentences
and send it to {recepient_email}. Include the link to the webpage into the body of the email."""
})

print(result)
> Entering new AgentExecutor chain...

Invoking: `CA72DFB0AB4DF6C830B43E14B0782F70` with `{'publicWebpageUrl': 'http://www.paulgraham.com/vb.html'}`


{'summary': 'The author reflects on the concept of life being short and how having children made them realize the true brevity of life. They discuss how time can be converted into discrete quantities and how limited certain experiences are. The author emphasizes the importance of prioritizing and eliminating unnecessary things in life, as well as actively pursuing meaningful experiences. They also discuss the negative impact of getting caught up in online arguments and the need to be aware of how time is being spent. The author suggests pruning unnecessary activities, not waiting to do things that matter, and savoring the time one has.'}
Invoking: `CABC80BB79C15067CA983495324AE709` with `{'recipient': 'test@example.com', 'subject': 'Summary of the webpage', 'body': 'Here is a short summary of the webpage http://www.paulgraham.com/vb.html:\n\nThe author reflects on the concept of life being short and how having children made them realize the true brevity of life. They discuss how time can be converted into discrete quantities and how limited certain experiences are. The author emphasizes the importance of prioritizing and eliminating unnecessary things in life, as well as actively pursuing meaningful experiences. They also discuss the negative impact of getting caught up in online arguments and the need to be aware of how time is being spent. The author suggests pruning unnecessary activities, not waiting to do things that matter, and savoring the time one has.\n\nYou can find the full webpage [here](http://www.paulgraham.com/vb.html).'}`


{'messageId': '[2f04b00e-122d-c7de-c91e-e78e0c3276d6@gmail.com](mailto:2f04b00e-122d-c7de-c91e-e78e0c3276d6@gmail.com)'}I have sent the email with the summary of the webpage to test@example.com. Please check your inbox.

> Finished chain.
I have sent the email with the summary of the webpage to test@example.com. Please check your inbox.
注意:Connery Action 是一个结构化工具,因此您只能在支持结构化工具的智能体中使用它。

工具

import os

from langchain.agents import create_agent
from langchain_community.tools.connery import ConneryService
from langchain_openai import ChatOpenAI

# Specify your Connery Runner credentials.
os.environ["CONNERY_RUNNER_URL"] = ""
os.environ["CONNERY_RUNNER_API_KEY"] = ""

# Specify OpenAI API key.
os.environ["OPENAI_API_KEY"] = ""

# Specify your email address to receive the emails from examples below.
recepient_email = "test@example.com"

# Get the SendEmail action from the Connery Runner by ID.
connery_service = ConneryService()
send_email_action = connery_service.get_action("CABC80BB79C15067CA983495324AE709")
手动运行操作。
manual_run_result = send_email_action.run(
    {
        "recipient": recepient_email,
        "subject": "Test email",
        "body": "This is a test email sent from Connery.",
    }
)
print(manual_run_result)
使用 OpenAI Functions 智能体运行操作。 您可以在这里看到此示例的 LangSmith 追踪。
llm = ChatOpenAI(temperature=0)

agent = create_agent(
    model=llm,
    tools=[send_email_action],
    verbose=True,
)
agent_run_result = agent.invoke({
    "input": f"Send an email to the {recepient_email} and say that I will be late for the meeting."
})

print(agent_run_result)
> Entering new AgentExecutor chain...

Invoking: `CABC80BB79C15067CA983495324AE709` with `{'recipient': 'test@example.com', 'subject': 'Late for Meeting', 'body': 'Dear Team,\n\nI wanted to inform you that I will be late for the meeting today. I apologize for any inconvenience caused. Please proceed with the meeting without me and I will join as soon as I can.\n\nBest regards,\n[Your Name]'}`


{'messageId': '[d34a694d-50e0-3988-25da-e86b4c51d7a7@gmail.com](mailto:d34a694d-50e0-3988-25da-e86b4c51d7a7@gmail.com)'}I have sent an email to test@example.com informing them that you will be late for the meeting.

> Finished chain.
I have sent an email to test@example.com informing them that you will be late for the meeting.
注意:Connery Action 是一个结构化工具,因此您只能在支持结构化工具的智能体中使用它。

API 参考

有关所有 Connery 功能和配置的详细文档,请访问 API 参考: