探索语义内核 OpenAIAssistantAgent

功能当前在 Java 中不可用。

将 Microsoft.SemanticKernel.Agents.OpenAI 包添加到项目：

dotnet add package Microsoft.SemanticKernel.Agents.OpenAI --prerelease

您可能还希望包括 Azure.Identity 包：

dotnet add package Azure.Identity

安装 semantic-kernel 包：

pip install semantic-kernel

功能当前在 Java 中不可用。

AssistantClient client = OpenAIAssistantAgent.CreateAzureOpenAIClient(...).GetAssistantClient();
Assistant assistant =
    await client.CreateAssistantAsync(
        "<model name>",
        "<agent name>",
        instructions: "<agent instructions>");
OpenAIAssistantAgent agent = new(assistant, client);

from semantic_kernel.agents import AssistantAgentThread, AzureAssistantAgent, OpenAIAssistantAgent
from semantic_kernel.connectors.ai.open_ai import AzureOpenAISettings, OpenAISettings

# Set up the client and model using Azure OpenAI Resources
client = AzureAssistantAgent.create_client()

# Define the assistant definition
definition = await client.beta.assistants.create(
    model=AzureOpenAISettings().chat_deployment_name,
    instructions="<instructions>",
    name="<agent name>",
)

# Create the AzureAssistantAgent instance using the client and the assistant definition
agent = AzureAssistantAgent(
    client=client,
    definition=definition,
)

# or

# Set up the client and model using OpenAI Resources
client = OpenAIAssistantAgent.create_client()

# Define the assistant definition
definition = await client.beta.assistants.create(
    model=OpenAISettings().chat_model_id,
    instructions="<instructions>",
    name="<agent name>",
)

# Create the OpenAIAssistantAgent instance using the client and the assistant definition
agent = OpenAIAssistantAgent(
    client=client,
    definition=definition,
)

功能当前在 Java 中不可用。

对于 .NET，代理标识符通过由任何代理定义的属性以 string 的形式公开。

AssistantClient client = OpenAIAssistantAgent.CreateAzureOpenAIClient(...).GetAssistantClient();
Assistant assistant = await client.GetAssistantAsync("<assistant id>");
OpenAIAssistantAgent agent = new(assistant, client);

# Using Azure OpenAI Resources

# Create the client using Azure OpenAI resources and configuration
client = AzureAssistantAgent.create_client()

# Create the assistant definition
definition = await client.beta.assistants.create(
    model=AzureOpenAISettings().chat_deployment_name,
    name="<agent name>",
    instructions="<instructions>",
)

# Store the assistant ID
assistant_id = definition.id

# Retrieve the assistant definition from the server based on the assistant ID
new_asst_definition = await client.beta.assistants.retrieve(assistant_id)

# Create the AzureAssistantAgent instance using the client and the assistant definition
agent = AzureAssistantAgent(
    client=client,
    definition=new_asst_definition,
)

功能当前在 Java 中不可用。

// Define agent
OpenAIAssistantAgent agent = ...;
AgentThread? agentThread = null;

// Generate the agent response(s)
await foreach (AgentResponseItem<ChatMessageContent> response in agent.InvokeAsync(new ChatMessageContent(AuthorRole.User, "<user input>")))
{
  // Process agent response(s)...
  agentThread = response.Thread;
}

// Delete the thread if no longer needed
if (agentThread is not null)
{
    await agentThread.DeleteAsync();
}

您还可以使用您创建的 OpenAIAssistantAgent 来调用 AgentThread。

// Define agent
OpenAIAssistantAgent agent = ...;

// Create a thread with some custom metadata.
AgentThread agentThread = new OpenAIAssistantAgentThread(client, metadata: myMetadata);

// Generate the agent response(s)
await foreach (ChatMessageContent response in agent.InvokeAsync(new ChatMessageContent(AuthorRole.User, "<user input>"), agentThread))
{
  // Process agent response(s)...
}

// Delete the thread when it is no longer needed
await agentThread.DeleteAsync();

您还可以创建一个通过对话编号恢复早期对话的 OpenAIAssistantAgentThread。

// Create a thread with an existing thread id.
AgentThread agentThread = new OpenAIAssistantAgentThread(client, "existing-thread-id");

from semantic_kernel.agents import AssistantAgentThread, AzureAssistantAgent

# Define agent
openai_agent = await ...

# Create a thread for the agent conversation
thread: AssistantAgentThread = None

# Generate the agent response(s)
async for response in agent.invoke(messages="user input", thread=thread):
  # process agent response(s)...
  thread = response.thread

# Delete the thread when it is no longer needed
await thread.delete() if thread else None

功能当前在 Java 中不可用。

对于 .NET，通过任何代理定义的string属性来公开代理标识符Agent.Id。

AssistantClient client = OpenAIAssistantAgent.CreateAzureOpenAIClient(...).GetAssistantClient();
await client.DeleteAssistantAsync("<assistant id>");

await client.beta.assistants.delete(agent.id)

功能当前在 Java 中不可用。

即将推出 OpenAIAssistantAgent 回调文档。

在on_intermediate_message或agent.invoke(...)中配置agent.invoke_stream(...)回调函数允许调用方在制定代理最终响应的过程中接收生成的中间消息。

import asyncio
from typing import Annotated

from semantic_kernel.agents import AssistantAgentThread, AzureAssistantAgent
from semantic_kernel.connectors.ai.open_ai import AzureOpenAISettings
from semantic_kernel.contents import AuthorRole, ChatMessageContent, FunctionCallContent, FunctionResultContent
from semantic_kernel.functions import kernel_function


# Define a sample plugin for the sample
class MenuPlugin:
    """A sample Menu Plugin used for the concept sample."""

    @kernel_function(description="Provides a list of specials from the menu.")
    def get_specials(self) -> Annotated[str, "Returns the specials from the menu."]:
        return """
        Special Soup: Clam Chowder
        Special Salad: Cobb Salad
        Special Drink: Chai Tea
        """

    @kernel_function(description="Provides the price of the requested menu item.")
    def get_item_price(
        self, menu_item: Annotated[str, "The name of the menu item."]
    ) -> Annotated[str, "Returns the price of the menu item."]:
        return "$9.99"


# This callback function will be called for each intermediate message,
# which will allow one to handle FunctionCallContent and FunctionResultContent.
# If the callback is not provided, the agent will return the final response
# with no intermediate tool call steps.
async def handle_intermediate_steps(message: ChatMessageContent) -> None:
    for item in message.items or []:
        if isinstance(item, FunctionResultContent):
            print(f"Function Result:> {item.result} for function: {item.name}")
        elif isinstance(item, FunctionCallContent):
            print(f"Function Call:> {item.name} with arguments: {item.arguments}")
        else:
            print(f"{item}")


async def main():
    # Create the client using Azure OpenAI resources and configuration
    client = AzureAssistantAgent.create_client()

    # Define the assistant definition
    definition = await client.beta.assistants.create(
        model=AzureOpenAISettings().chat_deployment_name,
        name="Host",
        instructions="Answer questions about the menu.",
    )

    # Create the AzureAssistantAgent instance using the client and the assistant definition and the defined plugin
    agent = AzureAssistantAgent(
        client=client,
        definition=definition,
        plugins=[MenuPlugin()],
    )

    # Create a new thread for use with the assistant
    # If no thread is provided, a new thread will be
    # created and returned with the initial response
    thread: AssistantAgentThread = None

    user_inputs = [
        "Hello",
        "What is the special soup?",
        "What is the special drink?",
        "How much is that?",
        "Thank you",
    ]

    try:
        for user_input in user_inputs:
            print(f"# {AuthorRole.USER}: '{user_input}'")
            async for response in agent.invoke(
                messages=user_input,
                thread=thread,
                on_intermediate_message=handle_intermediate_steps,
            ):
                print(f"# {response.role}: {response}")
                thread = response.thread
    finally:
        await thread.delete() if thread else None
        await client.beta.assistants.delete(assistant_id=agent.id)


if __name__ == "__main__":
    asyncio.run(main())

下面演示了代理调用过程的示例输出：

AuthorRole.USER: 'Hello'
AuthorRole.ASSISTANT: Hello! How can I assist you today?
AuthorRole.USER: 'What is the special soup?'
Function Call:> MenuPlugin-get_specials with arguments: {}
Function Result:> 
        Special Soup: Clam Chowder
        Special Salad: Cobb Salad
        Special Drink: Chai Tea
        for function: MenuPlugin-get_specials
AuthorRole.ASSISTANT: The special soup is Clam Chowder. Would you like to know more about the specials or 
    anything else?
AuthorRole.USER: 'What is the special drink?'
AuthorRole.ASSISTANT: The special drink is Chai Tea. If you have any more questions, feel free to ask!
AuthorRole.USER: 'How much is that?'
Function Call:> MenuPlugin-get_item_price with arguments: {"menu_item":"Chai Tea"}
Function Result:> $9.99 for function: MenuPlugin-get_item_price
AuthorRole.ASSISTANT: The Chai Tea is priced at $9.99. If there's anything else you'd like to know, 
    just let me know!
AuthorRole.USER: 'Thank you'
AuthorRole.ASSISTANT: You're welcome! If you have any more questions or need further assistance, feel free to 
    ask. Enjoy your day!

功能当前在 Java 中不可用。

即将推出有关使用声明式规格的文档。

OpenAIAssistantAgent 支持从 YAML 声明性规范中进行实例化。 声明性方法允许在单个可审核文档中定义代理的属性、说明、模型配置、工具和其他选项。 这使得代理组合可移植且易于跨环境进行管理。

如何使用声明性规范

本部分概述了关键原则，并提供概念示例的链接，这些示例显示了每种工具类型的完整代码，而不是枚举每个可能的 YAML 配置。 请参阅以下概念示例，了解具有声明性规范的 OpenAIAssistantAgent 端到端实现：

AzureAssistantAgent 样品：

• 代码解释器
• 文件搜索
• 文件中的函数插件
• 从现有助手 ID 加载
• 提示模板

OpenAIAssistantAgent 样品：

• 代码解释器
• 文件搜索
• 函数插件
• 文件中的函数插件
• 从现有助手 ID 加载
• 提示模板

示例：从 YAML 创建 AzureAIAgent

最小 YAML 声明性规范可能如下所示：

type: openai_assistant
name: Host
instructions: Respond politely to the user's questions.
model:
  id: ${OpenAI:ChatModelId}
tools:
  - id: MenuPlugin.get_specials
    type: function
  - id: MenuPlugin.get_item_price
    type: function

有关如何连接代理的详细信息，请参阅上面的完整代码示例。

要点

• 声明性规范允许在 YAML 中定义代理结构、工具和行为。
• 必须在运行时注册或访问所有引用的工具和插件。
• 必应、文件搜索和代码解释器等内置工具需要适当的配置和凭据（通常通过环境变量或显式参数）。
• 有关全面的示例，请参阅提供的示例链接，这些链接演示了实际方案，包括插件注册、Azure 标识配置和高级工具使用。

此功能不可用。