autogen_core.model_context#

class BufferedChatCompletionContext(buffer_size: int, initial_messages: List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]] | None = None)[source]#

Bases: ChatCompletionContext, Component[BufferedChatCompletionContextConfig]

A buffered chat completion context that keeps a view of the last n messages, where n is the buffer size. The buffer size is set at initialization.

Parameters:

buffer_size (int) – The size of the buffer.
initial_messages (List[LLMMessage] | None) – The initial messages.

classmethod _from_config(config: BufferedChatCompletionContextConfig) → Self[source]#

Create a new instance of the component from a configuration object.

Parameters:: config (T) – The configuration object.
Returns:: Self – The new instance of the component.

_to_config() → BufferedChatCompletionContextConfig[source]#

Dump the configuration that would be requite to create a new instance of a component matching the configuration of this instance.

Returns:: T – The configuration of the component.

component_config_schema#: alias of BufferedChatCompletionContextConfig

component_provider_override: ClassVar[str | None] = 'autogen_core.model_context.BufferedChatCompletionContext'#: Override the provider string for the component. This should be used to prevent internal module names being a part of the module name.

async get_messages() → List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]][source]#: Get at most buffer_size recent messages.

class ChatCompletionContext(initial_messages: List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]] | None = None)[source]#

Bases: ABC, ComponentBase[BaseModel]

An abstract base class for defining the interface of a chat completion context. A chat completion context lets agents store and retrieve LLM messages. It can be implemented with different recall strategies.

Parameters:: initial_messages (List[LLMMessage] | None) – The initial messages.

Example

To create a custom model context that filters out the thought field from AssistantMessage. This is useful for reasoning models like DeepSeek R1, which produces very long thought that is not needed for subsequent completions.

from typing import List

from autogen_core.model_context import UnboundedChatCompletionContext
from autogen_core.models import AssistantMessage, LLMMessage


class ReasoningModelContext(UnboundedChatCompletionContext):
    """A model context for reasoning models."""

    async def get_messages(self) -> List[LLMMessage]:
        messages = await super().get_messages()
        # Filter out thought field from AssistantMessage.
        messages_out: List[LLMMessage] = []
        for message in messages:
            if isinstance(message, AssistantMessage):
                message.thought = None
            messages_out.append(message)
        return messages_out

async add_message(message: Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]) → None[source]#: Add a message to the context.

async clear() → None[source]#: Clear the context.

component_type: ClassVar[ComponentType] = 'chat_completion_context'#: The logical type of the component.

abstract async get_messages() → List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]][source]#

async load_state(state: Mapping[str, Any]) → None[source]#

async save_state() → Mapping[str, Any][source]#

pydantic model ChatCompletionContextState[source]#

Bases: BaseModel

Show JSON schema

{
   "title": "ChatCompletionContextState",
   "type": "object",
   "properties": {
      "messages": {
         "items": {
            "discriminator": {
               "mapping": {
                  "AssistantMessage": "#/$defs/AssistantMessage",
                  "FunctionExecutionResultMessage": "#/$defs/FunctionExecutionResultMessage",
                  "SystemMessage": "#/$defs/SystemMessage",
                  "UserMessage": "#/$defs/UserMessage"
               },
               "propertyName": "type"
            },
            "oneOf": [
               {
                  "$ref": "#/$defs/SystemMessage"
               },
               {
                  "$ref": "#/$defs/UserMessage"
               },
               {
                  "$ref": "#/$defs/AssistantMessage"
               },
               {
                  "$ref": "#/$defs/FunctionExecutionResultMessage"
               }
            ]
         },
         "title": "Messages",
         "type": "array"
      }
   },
   "$defs": {
      "AssistantMessage": {
         "description": "Assistant message are sampled from the language model.",
         "properties": {
            "content": {
               "anyOf": [
                  {
                     "type": "string"
                  },
                  {
                     "items": {
                        "$ref": "#/$defs/FunctionCall"
                     },
                     "type": "array"
                  }
               ],
               "title": "Content"
            },
            "thought": {
               "anyOf": [
                  {
                     "type": "string"
                  },
                  {
                     "type": "null"
                  }
               ],
               "default": null,
               "title": "Thought"
            },
            "source": {
               "title": "Source",
               "type": "string"
            },
            "type": {
               "const": "AssistantMessage",
               "default": "AssistantMessage",
               "title": "Type",
               "type": "string"
            }
         },
         "required": [
            "content",
            "source"
         ],
         "title": "AssistantMessage",
         "type": "object"
      },
      "FunctionCall": {
         "properties": {
            "id": {
               "title": "Id",
               "type": "string"
            },
            "arguments": {
               "title": "Arguments",
               "type": "string"
            },
            "name": {
               "title": "Name",
               "type": "string"
            }
         },
         "required": [
            "id",
            "arguments",
            "name"
         ],
         "title": "FunctionCall",
         "type": "object"
      },
      "FunctionExecutionResult": {
         "description": "Function execution result contains the output of a function call.",
         "properties": {
            "content": {
               "title": "Content",
               "type": "string"
            },
            "name": {
               "title": "Name",
               "type": "string"
            },
            "call_id": {
               "title": "Call Id",
               "type": "string"
            },
            "is_error": {
               "anyOf": [
                  {
                     "type": "boolean"
                  },
                  {
                     "type": "null"
                  }
               ],
               "default": null,
               "title": "Is Error"
            }
         },
         "required": [
            "content",
            "name",
            "call_id"
         ],
         "title": "FunctionExecutionResult",
         "type": "object"
      },
      "FunctionExecutionResultMessage": {
         "description": "Function execution result message contains the output of multiple function calls.",
         "properties": {
            "content": {
               "items": {
                  "$ref": "#/$defs/FunctionExecutionResult"
               },
               "title": "Content",
               "type": "array"
            },
            "type": {
               "const": "FunctionExecutionResultMessage",
               "default": "FunctionExecutionResultMessage",
               "title": "Type",
               "type": "string"
            }
         },
         "required": [
            "content"
         ],
         "title": "FunctionExecutionResultMessage",
         "type": "object"
      },
      "SystemMessage": {
         "description": "System message contains instructions for the model coming from the developer.\n\n.. note::\n\n    Open AI is moving away from using 'system' role in favor of 'developer' role.\n    See `Model Spec <https://cdn.openai.com/spec/model-spec-2024-05-08.html#definitions>`_ for more details.\n    However, the 'system' role is still allowed in their API and will be automatically converted to 'developer' role\n    on the server side.\n    So, you can use `SystemMessage` for developer messages.",
         "properties": {
            "content": {
               "title": "Content",
               "type": "string"
            },
            "type": {
               "const": "SystemMessage",
               "default": "SystemMessage",
               "title": "Type",
               "type": "string"
            }
         },
         "required": [
            "content"
         ],
         "title": "SystemMessage",
         "type": "object"
      },
      "UserMessage": {
         "description": "User message contains input from end users, or a catch-all for data provided to the model.",
         "properties": {
            "content": {
               "anyOf": [
                  {
                     "type": "string"
                  },
                  {
                     "items": {
                        "anyOf": [
                           {
                              "type": "string"
                           },
                           {}
                        ]
                     },
                     "type": "array"
                  }
               ],
               "title": "Content"
            },
            "source": {
               "title": "Source",
               "type": "string"
            },
            "type": {
               "const": "UserMessage",
               "default": "UserMessage",
               "title": "Type",
               "type": "string"
            }
         },
         "required": [
            "content",
            "source"
         ],
         "title": "UserMessage",
         "type": "object"
      }
   }
}

Fields:

messages (List[autogen_core.models._types.SystemMessage | autogen_core.models._types.UserMessage | autogen_core.models._types.AssistantMessage | autogen_core.models._types.FunctionExecutionResultMessage])

field messages: List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]] [Optional]#

class HeadAndTailChatCompletionContext(head_size: int, tail_size: int, initial_messages: List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]] | None = None)[source]#

Bases: ChatCompletionContext, Component[HeadAndTailChatCompletionContextConfig]

A chat completion context that keeps a view of the first n and last m messages, where n is the head size and m is the tail size. The head and tail sizes are set at initialization.

Parameters:

head_size (int) – The size of the head.
tail_size (int) – The size of the tail.
initial_messages (List[LLMMessage] | None) – The initial messages.

classmethod _from_config(config: HeadAndTailChatCompletionContextConfig) → Self[source]#

Create a new instance of the component from a configuration object.

Parameters:: config (T) – The configuration object.
Returns:: Self – The new instance of the component.

_to_config() → HeadAndTailChatCompletionContextConfig[source]#

Dump the configuration that would be requite to create a new instance of a component matching the configuration of this instance.

Returns:: T – The configuration of the component.

component_config_schema#: alias of HeadAndTailChatCompletionContextConfig

component_provider_override: ClassVar[str | None] = 'autogen_core.model_context.HeadAndTailChatCompletionContext'#: Override the provider string for the component. This should be used to prevent internal module names being a part of the module name.

async get_messages() → List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]][source]#: Get at most head_size recent messages and tail_size oldest messages.

class TokenLimitedChatCompletionContext(model_client: ChatCompletionClient, *, token_limit: int | None = None, tool_schema: List[ToolSchema] | None = None, initial_messages: List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]] | None = None)[source]#

Bases: ChatCompletionContext, Component[TokenLimitedChatCompletionContextConfig]

(Experimental) A token based chat completion context maintains a view of the context up to a token limit.

Note

Added in v0.4.10. This is an experimental component and may change in the future.

Parameters:

model_client (ChatCompletionClient) – The model client to use for token counting. The model client must implement the count_tokens() and remaining_tokens() methods.
token_limit (int | None) – The maximum number of tokens to keep in the context using the count_tokens() method. If None, the context will be limited by the model client using the remaining_tokens() method.
tools (List[ToolSchema] | None) – A list of tool schema to use in the context.
initial_messages (List[LLMMessage] | None) – A list of initial messages to include in the context.

classmethod _from_config(config: TokenLimitedChatCompletionContextConfig) → Self[source]#

Create a new instance of the component from a configuration object.

Parameters:: config (T) – The configuration object.
Returns:: Self – The new instance of the component.

_to_config() → TokenLimitedChatCompletionContextConfig[source]#

Dump the configuration that would be requite to create a new instance of a component matching the configuration of this instance.

Returns:: T – The configuration of the component.

component_config_schema#: alias of TokenLimitedChatCompletionContextConfig

component_provider_override: ClassVar[str | None] = 'autogen_core.model_context.TokenLimitedChatCompletionContext'#: Override the provider string for the component. This should be used to prevent internal module names being a part of the module name.

async get_messages() → List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]][source]#: Get at most token_limit tokens in recent messages. If the token limit is not provided, then return as many messages as the remaining token allowed by the model client.

class UnboundedChatCompletionContext(initial_messages: List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]] | None = None)[source]#

Bases: ChatCompletionContext, Component[UnboundedChatCompletionContextConfig]

An unbounded chat completion context that keeps a view of the all the messages.

classmethod _from_config(config: UnboundedChatCompletionContextConfig) → Self[source]#

Create a new instance of the component from a configuration object.

Parameters:: config (T) – The configuration object.
Returns:: Self – The new instance of the component.

_to_config() → UnboundedChatCompletionContextConfig[source]#

Dump the configuration that would be requite to create a new instance of a component matching the configuration of this instance.

Returns:: T – The configuration of the component.

component_config_schema#: alias of UnboundedChatCompletionContextConfig

component_provider_override: ClassVar[str | None] = 'autogen_core.model_context.UnboundedChatCompletionContext'#: Override the provider string for the component. This should be used to prevent internal module names being a part of the module name.

async get_messages() → List[Annotated[SystemMessage | UserMessage | AssistantMessage | FunctionExecutionResultMessage, FieldInfo(annotation=NoneType, required=True, discriminator='type')]][source]#: Get at most buffer_size recent messages.