Share via

Agent Class

Base abstraction for all Semantic Kernel agents. An agent instance may participate in one or more conversations. A conversation may include one or more agents. In addition to identity and descriptive meta-data, an Agent must define its communication protocol, or AgentChannel. Create a new model by parsing and validating input data from keyword arguments. Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model. self is explicitly positional-only to allow self as a field name.

Constructor

Agent(
  *,
  arguments: KernelArguments | None = None,
  description: str | None = None,
  id: str | None = None,
  instructions: str | None = None,
  kernel: Kernel | None = None,
  name: Annotated[str, _PydanticGeneralMetadata(pattern='^[0-9A-Za-z_-]+$')] | None = None,
  prompt_template: PromptTemplateBase | None = None
)

Keyword-Only Parameters

Name Description
arguments
description
id
instructions
kernel
name
prompt_template

Methods

create_channel

Create a channel.
format_instructions

Format the instructions.
get_channel_keys

Get the channel keys.
get_response

Get a response from the agent. This method returns the final result of the agent's execution as an AgentResponseItem containing a single ChatMessageContent object. The caller is blocked until the final result is available. Note: For streaming responses, use the invoke_stream method, which returns intermediate steps and the final result as a stream of AgentResponseItem[ StreamingChatMessageContent]. Streaming only the final result is not feasible because the timing of the final result’s availability is unknown, and blocking the caller until then is undesirable in streaming scenarios.
invoke

Invoke the agent. This invocation method will return the intermediate steps and the final results of the agent's execution as an asynchronous stream of AgentResponseItem[ChatMessageContent] objects to the caller. Note: A ChatMessageContent object contains an entire message.
invoke_stream

Invoke the agent as a stream. This invocation method will return the intermediate steps and final results of the agent's execution as an asynchronous stream of AgentResponseItem[StreamingChatMessageContent] objects to the caller. Note: A StreamingChatMessageContent object contains a chunk of a message.

create_channel

Create a channel.

async create_channel() -> AgentChannel

Returns

Type Description

An instance of AgentChannel.

format_instructions

Format the instructions.

async format_instructions(
  kernel: Kernel,
  arguments: KernelArguments | None = None
) -> str | None

Parameters

Name Description
kernel
Required

The kernel instance.
arguments

The kernel arguments.

Default value: None

Returns

Type Description

The formatted instructions.

get_channel_keys

Get the channel keys.

get_channel_keys() -> Iterable[str]

Returns

Type Description

A list of channel keys.

get_response

Get a response from the agent. This method returns the final result of the agent's execution as an AgentResponseItem containing a single ChatMessageContent object. The caller is blocked until the final result is available. Note: For streaming responses, use the invoke_stream method, which returns intermediate steps and the final result as a stream of AgentResponseItem[ StreamingChatMessageContent]. Streaming only the final result is not feasible because the timing of the final result’s availability is unknown, and blocking the caller until then is undesirable in streaming scenarios.

abstract get_response(
  messages: str | ChatMessageContent | list[str | ChatMessageContent] | None = None,
  thread: AgentThread | None = None,
  **kwargs
) -> Awaitable[AgentResponseItem[ChatMessageContent]]

Parameters

Name Description
messages

The message(s) to send to the agent.

Default value: None
thread

The conversation thread associated with the message(s).

Default value: None

Returns

Type Description

An AgentResponseItem containing the final ChatMessageContent.

invoke

Invoke the agent. This invocation method will return the intermediate steps and the final results of the agent's execution as an asynchronous stream of AgentResponseItem[ChatMessageContent] objects to the caller. Note: A ChatMessageContent object contains an entire message.

abstract invoke(
  messages: str | ChatMessageContent | list[str | ChatMessageContent] | None = None,
  thread: AgentThread | None = None,
  **kwargs
) -> AsyncIterable[AgentResponseItem[ChatMessageContent]]

Parameters

Name Description
messages

The message(s) to send to the agent.

Default value: None
thread

The conversation thread associated with the message(s).

Default value: None

Returns

Type Description

An asynchronous stream of AgentResponseItem containing ChatMessageContent.

invoke_stream

Invoke the agent as a stream. This invocation method will return the intermediate steps and final results of the agent's execution as an asynchronous stream of AgentResponseItem[StreamingChatMessageContent] objects to the caller. Note: A StreamingChatMessageContent object contains a chunk of a message.

abstract invoke_stream(
  messages: str | ChatMessageContent | list[str | ChatMessageContent] | None = None,
  thread: AgentThread | None = None,
  **kwargs
) -> AsyncIterable[AgentResponseItem[StreamingChatMessageContent]]

Parameters

Name Description
messages

The message(s) to send to the agent.

Default value: None
thread

The conversation thread associated with the message(s).

Default value: None

Returns

Type Description

An asynchronous stream of AgentResponseItem containing StreamingChatMessageContent.

Attributes

arguments

The arguments for the agent

arguments: KernelArguments | None

channel_type

The type of the agent channel

channel_type: ClassVar[type[AgentChannel] | None] = None

description

The description of the agent

description: str | None

id

The unique identifier of the agent. If no id is provided, a new UUID will be generated.

id: str

instructions

The instructions for the agent (optional)

instructions: str | None

kernel

The kernel instance for the agent

kernel: Kernel

name

The name of the agent

name: str

prompt_template

The prompt template for the agent

prompt_template: PromptTemplateBase | None