🤖 Agents ¶

CustomizeAgent provides a flexible framework for creating specialized LLM-powered agents without writing custom code. It enables the creation of agents with well-defined inputs and outputs, custom prompt templates, and configurable parsing strategies.

Attributes:

Name	Type	Description
`name`	`str`	The name of the agent.
`description`	`str`	A description of the agent's purpose and capabilities.
`prompt_template`	`PromptTemplate`	The prompt template that will be used for the agent's primary action.
`prompt`	`str`	The prompt template that will be used for the agent's primary action. Should contain placeholders in the format `{input_name}` for each input parameter.
`llm_config`	`LLMConfig`	Configuration for the language model.
`inputs`	`List[Union[dict, Parameter]]`	List of input specifications as dicts or Parameter objects. Each dict (e.g., `{"name": str, "type": str, "description": str, ["required": bool, "json_schema": dict]}`) contains: - name (str): Name of the input parameter - type (str): Type of the input - description (str): Description of what the input represents - required (bool, optional): Whether this input is required (default: True) - json_schema (dict, optional): The json schema of the input, recommended when type is `object` or `array`.
`outputs`	`List[Union[dict, Parameter]]`	List of output specifications as dicts or Parameter objects. Each dict (e.g., `{"name": str, "type": str, "description": str, ["required": bool, "json_schema": dict]}`) contains: - name (str): Name of the output field - type (str): Type of the output - description (str): Description of what the output represents - required (bool, optional): Whether this output is required (default: True) - json_schema (dict, optional): The json schema of the output, recommended when type is `object` or `array`.
`system_prompt`	`str`	The system prompt for the LLM. Defaults to DEFAULT_SYSTEM_PROMPT.
`output_parser`	`Type[ActionOutput]`	A custom class for parsing the LLM's output. Must be a subclass of ActionOutput.
`parse_mode`	`str`	Mode for parsing LLM output. Options are: - "title": Parse outputs using section titles (default) - "str": Parse as plain text - "json": Parse as JSON - "xml": Parse as XML - "custom": Use a custom parsing function
`parse_func`	`Callable`	Custom function for parsing LLM output when parse_mode is "custom". Must accept a "content" parameter and return a dictionary.
`title_format`	`str`	Format string for title parsing mode with {title} placeholder. Default is "## {title}".
`tools`	`list[Toolkit]`	List of tools to be used by the agent.
`custom_output_format`	`str`	Specify the output format. Only used when `prompt_template` is used. If not provided, the output format will be constructed from the `outputs` specification and `parse_mode`.

Source code in evoagentx/agents/customize_agent.py

def __init__(
    self,
    name: str,
    description: str,
    prompt: Optional[str] = None,
    prompt_template: Optional[PromptTemplate] = None,
    llm_config: Optional[LLMConfig] = None,
    inputs: Optional[List[Union[dict, Parameter]]] = None,
    outputs: Optional[List[Union[dict, Parameter]]] = None,
    system_prompt: Optional[str] = None,
    output_parser: Optional[Type[ActionOutput]] = None,
    parse_mode: Optional[str] = "title",
    parse_func: Optional[Callable] = None,
    title_format: Optional[str] = None,
    tools: Optional[List[Union[Toolkit, Tool]]] = None,
    custom_output_format: Optional[str] = None,
    max_steps: int = 20,
    max_tool_call_concurrency: int = 5,
    **kwargs
):
    system_prompt = system_prompt or DEFAULT_SYSTEM_PROMPT
    inputs = inputs or []
    outputs = outputs or []

    if prompt is not None and prompt_template is not None:
        logger.warning("Both `prompt` and `prompt_template` are provided in `CustomizeAgent`. `prompt_template` will be used.")
        prompt = None

    if isinstance(parse_func, str):
        if not PARSE_FUNCTION_REGISTRY.has_function(parse_func):
            raise ValueError(f"parse function `{parse_func}` is not registered! To instantiate a CustomizeAgent from a file, you should use decorator `@register_parse_function` to register the parse function.")
        parse_func = PARSE_FUNCTION_REGISTRY.get_function(parse_func)

    if isinstance(output_parser, str):
        output_parser = MODULE_REGISTRY.get_module(output_parser)

    # set default title format
    if parse_mode == "title" and title_format is None:
        title_format = "## {title}"

    # validate the data and normalize inputs/outputs to Parameter objects
    valid_inputs, valid_outputs, parse_mode = self.validate_data(
        prompt=prompt,
        prompt_template=prompt_template,
        inputs=inputs,
        outputs=outputs,
        output_parser=output_parser,
        parse_mode=parse_mode,
        parse_func=parse_func,
        title_format=title_format
    )

    customize_action = CustomizeAgent.create_customize_action(
        name=name,
        desc=description,
        prompt=prompt,
        prompt_template=prompt_template,
        inputs=valid_inputs,
        outputs=valid_outputs,
        parse_mode=parse_mode,
        parse_func=parse_func,
        output_parser=output_parser,
        title_format=title_format,
        custom_output_format=custom_output_format,
        tools=tools,
        max_steps=max_steps,
        max_tool_call_concurrency=max_tool_call_concurrency,
    )
    super().__init__(
        name=name,
        description=description,
        llm_config=llm_config,
        system_prompt=system_prompt,
        actions=[customize_action],
        **kwargs
    )

    # Set backing attributes after super().__init__ so Pydantic doesn't wipe them.
    # parse_func must be stored before parse_mode because the parse_mode setter reads it.
    self._inputs  = valid_inputs
    self._outputs = valid_outputs
    self.output_parser = output_parser
    self._parse_func = parse_func
    self.parse_mode = parse_mode
    self._title_format = title_format
    self.custom_output_format = custom_output_format

customize_action_name `property` ¶

customize_action_name: str

Get the name of the primary custom action for this agent.

Returns:

Type	Description
`str`	The name of the primary custom action

action `property` ¶

action: Action

Get the primary custom action for this agent.

Returns:

Type	Description
`Action`	The primary custom action

prompt `property` ¶

prompt: str

Get the prompt for the primary custom action.

Returns:

Type	Description
`str`	The prompt for the primary custom action

prompt_template `property` ¶

prompt_template: PromptTemplate

Get the prompt template for the primary custom action.

Returns:

Type	Description
`PromptTemplate`	The prompt template for the primary custom action

validate_data ¶

validate_data(prompt: str, prompt_template: PromptTemplate, inputs: List[Union[dict, Parameter]], outputs: List[Union[dict, Parameter]], output_parser: Type[ActionOutput], parse_mode: str, parse_func: Callable, title_format: str) -> tuple

Validate and normalize agent configuration, auto-correcting parse_mode where needed.

Converts inputs and outputs to Parameter objects, validates all parsing-related options, and (for prompt_template-based agents only) auto-corrects parse_mode to "json" when the output schema contains object/array types or a json_schema without a custom parse function. Raw-prompt agents keep their parse_mode, since the prompt is sent verbatim and dictates its own output format.

Returns:

Type	Description
`tuple`	A tuple containing:
`tuple`	`valid_inputs` (List[Parameter]): normalized input parameters.
`tuple`	`valid_outputs` (List[Parameter]): normalized output parameters.
`tuple`	`parse_mode` (str): validated (and possibly auto-corrected) parse mode.

Source code in evoagentx/agents/customize_agent.py

def validate_data(
    self,
    prompt: str,
    prompt_template: PromptTemplate,
    inputs: List[Union[dict, Parameter]],
    outputs: List[Union[dict, Parameter]],
    output_parser: Type[ActionOutput],
    parse_mode: str,
    parse_func: Callable,
    title_format: str,
) -> tuple:
    """Validate and normalize agent configuration, auto-correcting parse_mode where needed.

    Converts `inputs` and `outputs` to `Parameter` objects, validates all
    parsing-related options, and (for `prompt_template`-based agents only)
    auto-corrects `parse_mode` to `"json"` when the output schema contains
    `object`/`array` types or a `json_schema` without a custom parse function.
    Raw-`prompt` agents keep their `parse_mode`, since the prompt is sent verbatim
    and dictates its own output format.

    Returns:
        A tuple containing:
        - `valid_inputs` (List[Parameter]): normalized input parameters.
        - `valid_outputs` (List[Parameter]): normalized output parameters.
        - `parse_mode` (str): validated (and possibly auto-corrected) parse mode.
    """
    # Normalize inputs and outputs to Parameter objects first so all subsequent code
    # can safely use attribute access regardless of what the caller passed in.
    valid_inputs  = self._check_params_types(inputs,  "inputs")  or []
    valid_outputs = self._check_params_types(outputs, "outputs") or []

    # check if the prompt is provided
    if prompt is None and prompt_template is None:
        raise ValueError("`prompt` or `prompt_template` is required when creating a CustomizeAgent.")

    # check if all the inputs are in the prompt (only used when prompt_template is not provided)
    if prompt_template is None and valid_inputs:
        all_input_names = [input_item.name for input_item in valid_inputs]
        inputs_names_not_in_prompt = [name for name in all_input_names if f'{{{name}}}' not in prompt]
        if inputs_names_not_in_prompt:
            raise KeyError(f"The following inputs are not found in the prompt: {inputs_names_not_in_prompt}.")

    # check if the output_parser is valid
    if output_parser is not None:
        self._check_output_parser(valid_outputs, output_parser)

    # check the parse_mode value itself
    if parse_mode not in PARSER_VALID_MODE:
        raise ValueError(f"'{parse_mode}' is an invalid value for `parse_mode`. Available choices: {PARSER_VALID_MODE}.")

    # Auto-correct parse_mode to "json" when outputs require it and no custom parse_func is provided.
    # Only do this for prompt_template-based agents: the template renders the output-format section
    # (and injects the JSON schema) so the model is actually instructed to emit JSON. A raw `prompt`
    # is sent verbatim, so the model follows whatever format the prompt itself specifies; forcing
    # json here would make the parser disagree with the prompt's requested format.
    if prompt_template is not None and CustomizeAgent._outputs_require_json_mode(valid_outputs, parse_func) and parse_mode != "json":
        logger.warning(
            f"parse_mode='{parse_mode}' is not compatible with the current outputs (object/array types or "
            f"json_schema). Auto-correcting to parse_mode='json'. To suppress this warning, explicitly set "
            f"parse_mode='json'."
        )
        return valid_inputs, valid_outputs, "json"

    if parse_mode == "custom" and parse_func is None:
        raise ValueError("`parse_func` (a callable function with an input argument `content`) must be provided when `parse_mode` is 'custom'.")

    CustomizeAgent._validate_parse_func(parse_func)
    CustomizeAgent._validate_title_format(title_format, parse_mode)

    return valid_inputs, valid_outputs, parse_mode

create_customize_action `staticmethod` ¶

create_customize_action(name: str, desc: str, prompt: str, prompt_template: PromptTemplate, inputs: List[Union[dict, Parameter]], outputs: List[Union[dict, Parameter]], parse_mode: str, parse_func: Optional[Callable] = None, output_parser: Optional[ActionOutput] = None, title_format: Optional[str] = '## {title}', custom_output_format: Optional[str] = None, tools: Optional[List[Union[Tool, Toolkit]]] = None, max_steps: int = 20, max_tool_call_concurrency: int = 5, **kwargs) -> Action

Create a custom action based on the provided specifications.

This method dynamically generates an Action class and instance with: - Input parameters defined by the inputs specification - Output format defined by the outputs specification - Custom execution logic using the customize_action_execute function - If tools is provided, returns a CustomizeAction action instead

Parameters:

Name	Type	Description	Default
`name`	`str`	Base name for the action	required
`desc`	`str`	Description of the action	required
`prompt`	`str`	Prompt template for the action	required
`prompt_template`	`PromptTemplate`	Prompt template for the action	required
`inputs`	`List[Union[dict, Parameter]]`	List of input field specifications	required
`outputs`	`List[Union[dict, Parameter]]`	List of output field specifications	required
`parse_mode`	`str`	Mode to use for parsing LLM output	required
`parse_func`	`Optional[Callable]`	Optional custom parsing function	`None`
`output_parser`	`Optional[ActionOutput]`	Optional custom output parser class	`None`
`tools`	`Optional[List[Union[Tool, Toolkit]]]`	Optional list of tools	`None`
`max_steps`	`int`	Maximum number of steps the agent can take	`20`
`max_tool_call_concurrency`	`int`	Maximum number of concurrent tool calls	`5`

Returns:

Type	Description
`Action`	A newly created Action instance

Source code in evoagentx/agents/customize_agent.py

@staticmethod
def create_customize_action(
    name: str,
    desc: str,
    prompt: str,
    prompt_template: PromptTemplate,
    inputs: List[Union[dict, Parameter]],
    outputs: List[Union[dict, Parameter]],
    parse_mode: str,
    parse_func: Optional[Callable] = None,
    output_parser: Optional[ActionOutput] = None,
    title_format: Optional[str] = "## {title}",
    custom_output_format: Optional[str] = None,
    tools: Optional[List[Union[Tool, Toolkit]]] = None,
    max_steps: int = 20,
    max_tool_call_concurrency: int = 5,
    **kwargs
) -> Action:
    """Create a custom action based on the provided specifications.

    This method dynamically generates an Action class and instance with:
    - Input parameters defined by the inputs specification
    - Output format defined by the outputs specification
    - Custom execution logic using the customize_action_execute function
    - If tools is provided, returns a CustomizeAction action instead

    Args:
        name: Base name for the action
        desc: Description of the action
        prompt: Prompt template for the action
        prompt_template: Prompt template for the action
        inputs: List of input field specifications
        outputs: List of output field specifications
        parse_mode: Mode to use for parsing LLM output
        parse_func: Optional custom parsing function
        output_parser: Optional custom output parser class
        tools: Optional list of tools
        max_steps: Maximum number of steps the agent can take
        max_tool_call_concurrency: Maximum number of concurrent tool calls

    Returns:
        A newly created Action instance
    """
    assert prompt is not None or prompt_template is not None, "must provide `prompt` or `prompt_template` when creating CustomizeAgent"

    inputs: List[Parameter] = to_params(inputs)
    outputs: List[Parameter] = to_params(outputs)

    action_input_type = CustomizeAgent.create_action_input(inputs, name)

    if output_parser is None:
        action_output_type = CustomizeAgent.create_action_output(outputs, name)
    else:
        action_output_type = output_parser

    action_cls_name = get_unique_class_name(
        generate_dynamic_class_name(name + " action")
    )

    customize_action = CustomizeAction(
        name=action_cls_name,
        description=desc,
        prompt=prompt,
        prompt_template=prompt_template,
        inputs_format=action_input_type,
        outputs_format=action_output_type,
        parse_mode=parse_mode,
        parse_func=parse_func,
        title_format=title_format,
        custom_output_format=custom_output_format,
        tools=tools,
        max_steps=max_steps,
        max_tool_call_concurrency=max_tool_call_concurrency,
    )

    return customize_action

call ¶

__call__(inputs: dict = None, return_msg_type: MessageType = UNKNOWN, **kwargs) -> Message

Call the customize action.

Parameters:

Name	Type	Description	Default
`inputs`	`dict`	The inputs to the customize action.	`None`
`**kwargs`	`Any`	Additional keyword arguments.	`{}`

Returns:

Name	Type	Description
`ActionOutput`	`Message`	The output of the customize action.

Source code in evoagentx/agents/customize_agent.py

def __call__(self, inputs: dict = None, return_msg_type: MessageType = MessageType.UNKNOWN, **kwargs) -> Message:
    """
    Call the customize action.

    Args:
        inputs (dict): The inputs to the customize action.
        **kwargs (Any): Additional keyword arguments.

    Returns:
        ActionOutput: The output of the customize action.
    """
    # return self.execute(action_name=self.customize_action_name, action_input_data=inputs, **kwargs) 
    inputs = inputs or {} 
    return super().__call__(action_name=self.customize_action_name, action_input_data=inputs, return_msg_type=return_msg_type, **kwargs)

get_customize_agent_info ¶

get_customize_agent_info() -> dict

Get the information of the customize agent.

Source code in evoagentx/agents/customize_agent.py

def get_customize_agent_info(self) -> dict:
    """
    Get the information of the customize agent.
    """
    customize_action = self.get_action(self.customize_action_name)

    config = {
        "class_name": "CustomizeAgent",
        "name": self.name,
        "description": self.description,
        "prompt": customize_action.prompt,
        "prompt_template": customize_action.prompt_template.to_dict() if customize_action.prompt_template is not None else None, 
        # "llm_config": self.llm_config.to_dict(exclude_none=True),
        "inputs": [p.to_dict(ignore=["class_name"]) for p in self.inputs] if self.inputs else [],
        "outputs": [p.to_dict(ignore=["class_name"]) for p in self.outputs] if self.outputs else [],
        "system_prompt": self.system_prompt,
        "output_parser": self.output_parser.__name__ if self.output_parser is not None else None,
        "parse_mode": self.parse_mode,
        "parse_func": self.parse_func.__name__ if self.parse_func is not None else None,
        "title_format": self.title_format,
        "tool_names": [tool.name for tool in customize_action.tools] if customize_action.tools else [],
        "custom_output_format": self.custom_output_format,
        "max_steps": customize_action.max_steps,
        "max_tool_call_concurrency": customize_action.max_tool_call_concurrency,
    }
    return config

save_module ¶

save_module(path: str, ignore: List[str] = [], **kwargs) -> str

Save the customize agent's configuration to a JSON file.

Parameters:

Name	Type	Description	Default
`path`	`str`	File path where the configuration should be saved	required
`ignore`	`List[str]`	List of keys to exclude from the saved configuration	`[]`
`**kwargs`	`Any`	Additional parameters for the save operation	`{}`

Returns:

Type	Description
`str`	The path where the configuration was saved

Source code in evoagentx/agents/customize_agent.py

def save_module(self, path: str, ignore: List[str] = [], **kwargs)-> str:
    """Save the customize agent's configuration to a JSON file.

    Args:
        path: File path where the configuration should be saved
        ignore: List of keys to exclude from the saved configuration
        **kwargs (Any): Additional parameters for the save operation

    Returns:
        The path where the configuration was saved
    """
    config = self.get_customize_agent_info()

    for ignore_key in ignore:
        config.pop(ignore_key, None)

    # Save to JSON file
    make_parent_folder(path)
    with open(path, 'w', encoding='utf-8') as f:
        json.dump(config, f, indent=4, ensure_ascii=False)

    return path

get_config ¶

get_config() -> dict

Get a dictionary containing all necessary configuration to recreate this agent.

Returns:

Name	Type	Description
`dict`	`dict`	A configuration dictionary that can be used to initialize a new Agent instance
	`dict`	with the same properties as this one.

Source code in evoagentx/agents/customize_agent.py

def get_config(self) -> dict:
    """
    Get a dictionary containing all necessary configuration to recreate this agent.

    Returns:
        dict: A configuration dictionary that can be used to initialize a new Agent instance
        with the same properties as this one.
    """
    config = self.get_customize_agent_info()
    config["llm_config"] = self.llm_config.to_dict()
    return config

ActionAgent ¶

ActionAgent(name: str, description: str, inputs: List[dict], outputs: List[dict], execute_func: Callable, async_execute_func: Optional[Callable] = None, llm_config: Optional[LLMConfig] = None, **kwargs)

Bases: Agent

ActionAgent is a specialized agent that executes a provided function directly without LLM. It creates an action that uses the provided function as the execution backbone.

Attributes:

Name	Type	Description
`name`	`str`	The name of the agent.
`description`	`str`	A description of the agent's purpose and capabilities.
`inputs`	`List[dict]`	List of input specifications, where each dict contains: - name (str): Name of the input parameter - type (str): Type of the input - description (str): Description of what the input represents - required (bool, optional): Whether this input is required (default: True)
`outputs`	`List[dict]`	List of output specifications, where each dict contains: - name (str): Name of the output field - type (str): Type of the output - description (str): Description of what the output represents - required (bool, optional): Whether this output is required (default: True)
`execute_func`	`Callable`	The function to execute the agent.
`async_execute_func`	`(Callable, Optional)`	Async version of the function. If not provided, an async wrapper will be automatically created around execute_func.
`llm_config`	`LLMConfig`	Configuration for the language model (minimal usage).

Source code in evoagentx/agents/action_agent.py

def __init__(
    self,
    name: str,
    description: str,
    inputs: List[dict],
    outputs: List[dict],
    execute_func: Callable,
    async_execute_func: Optional[Callable] = None,
    llm_config: Optional[LLMConfig] = None,
    **kwargs
):
    # Validate inputs
    if not callable(execute_func):
        raise ValueError("execute_func must be callable")

    if async_execute_func is not None and not callable(async_execute_func):
        raise ValueError("async_execute_func must be callable")

    # Validate inputs and outputs
    self._validate_inputs_outputs(inputs, outputs)

    # Set is_human based on LLM availability
    is_human = llm_config is None

    # Initialize parent directly
    super().__init__(
        name=name,
        description=description,
        llm_config=llm_config,
        is_human=is_human,
        **kwargs
    )

    # Store function references and metadata
    self.execute_func = execute_func
    self.async_execute_func = async_execute_func
    self.inputs = inputs
    self.outputs = outputs

    # Create and add the function-based action
    action = self._create_function_action_with_params(
        name, execute_func, async_execute_func, inputs, outputs
    )
    self.add_action(action)

main_action_name `property` ¶

main_action_name: str

Get the name of the main function action for this agent.

Returns:

Type	Description
`str`	The name of the main function action

get_config ¶

get_config() -> dict

Get configuration for the ActionAgent.

Source code in evoagentx/agents/action_agent.py

def get_config(self) -> dict:
    """Get configuration for the ActionAgent."""
    # Get base config from Agent
    config = super().get_config()

    # Add ActionAgent-specific information
    config.update({
        "class_name": "ActionAgent",
        "execute_func_name": self.execute_func.__name__ if self.execute_func else None,
        "async_execute_func_name": self.async_execute_func.__name__ if self.async_execute_func else None,
        "inputs": self.inputs,
        "outputs": self.outputs
    })
    return config

save_module ¶

save_module(path: str, ignore: List[str] = [], **kwargs) -> str

Save the ActionAgent configuration to a JSON file.

Parameters:

Name	Type	Description	Default
`path`	`str`	File path where the configuration should be saved	required
`ignore`	`List[str]`	List of keys to exclude from the saved configuration	`[]`
`**kwargs`	`Any`	Additional parameters for the save operation	`{}`

Returns:

Type	Description
`str`	The path where the configuration was saved

Source code in evoagentx/agents/action_agent.py

def save_module(self, path: str, ignore: List[str] = [], **kwargs) -> str:
    """Save the ActionAgent configuration to a JSON file.

    Args:
        path: File path where the configuration should be saved
        ignore: List of keys to exclude from the saved configuration
        **kwargs (Any): Additional parameters for the save operation

    Returns:
        The path where the configuration was saved
    """
    config = self.get_config()

    # Add ActionAgent-specific information
    config.update({
        "class_name": "ActionAgent",
        "execute_func_name": self.execute_func.__name__ if self.execute_func else None,
        "async_execute_func_name": self.async_execute_func.__name__ if self.async_execute_func else None,
        "inputs": self.inputs,
        "outputs": self.outputs
    })

    # Remove non-serializable items
    for ignore_key in ignore:
        config.pop(ignore_key, None)

    # Save to JSON file
    make_parent_folder(path)
    with open(path, 'w', encoding='utf-8') as f:
        json.dump(config, f, indent=4, ensure_ascii=False)

    return path

load_module `classmethod` ¶

load_module(path: str, llm_config: LLMConfig = None, **kwargs) -> ActionAgent

Load the ActionAgent from a JSON file.

Parameters:

Name	Type	Description	Default
`path`	`str`	The path of the file	required
`llm_config`	`LLMConfig`	The LLMConfig instance (optional)	`None`
`**kwargs`		Additional keyword arguments	`{}`

Returns:

Name	Type	Description
`ActionAgent`	`ActionAgent`	The loaded agent instance

Raises:

Type	Description
`KeyError`	If required functions are not found in the registry

Source code in evoagentx/agents/action_agent.py

@classmethod
def load_module(cls, path: str, llm_config: LLMConfig = None, **kwargs) -> "ActionAgent":
    """Load the ActionAgent from a JSON file.

    Args:
        path: The path of the file
        llm_config: The LLMConfig instance (optional)
        **kwargs: Additional keyword arguments

    Returns:
        ActionAgent: The loaded agent instance

    Raises:
        KeyError: If required functions are not found in the registry
    """
    # Load configuration
    with open(path, 'r', encoding='utf-8') as f:
        config = json.load(f)

    # Extract function names
    execute_func_name = config.get("execute_func_name")
    async_execute_func_name = config.get("async_execute_func_name")

    # Retrieve functions from registry
    execute_func = None
    async_execute_func = None

    if execute_func_name:
        if not ACTION_FUNCTION_REGISTRY.has_function(execute_func_name):
            raise KeyError(f"Function '{execute_func_name}' not found in registry. Please register it first.")
        execute_func = ACTION_FUNCTION_REGISTRY.get_function(execute_func_name)

    if async_execute_func_name:
        if not ACTION_FUNCTION_REGISTRY.has_function(async_execute_func_name):
            raise KeyError(f"Function '{async_execute_func_name}' not found in registry. Please register it first.")
        async_execute_func = ACTION_FUNCTION_REGISTRY.get_function(async_execute_func_name)

    # Create agent
    agent = cls(
        name=config["name"],
        description=config["description"],
        inputs=config["inputs"],
        outputs=config["outputs"],
        execute_func=execute_func,
        async_execute_func=async_execute_func,
        llm_config=llm_config,
        **kwargs
    )

    return agent

call ¶

__call__(inputs: dict = None, return_msg_type: MessageType = UNKNOWN, **kwargs) -> Message

Call the main function action.

Parameters:

Name	Type	Description	Default
`inputs`	`dict`	The inputs to the function action.	`None`
`return_msg_type`	`MessageType`	The type of message to return.	`UNKNOWN`
`**kwargs`	`Any`	Additional keyword arguments.	`{}`

Returns:

Name	Type	Description
`Message`	`Message`	The output of the function action.

Source code in evoagentx/agents/action_agent.py

def __call__(self, inputs: dict = None, return_msg_type: MessageType = MessageType.UNKNOWN, **kwargs) -> Message:
    """
    Call the main function action.

    Args:
        inputs (dict): The inputs to the function action.
        return_msg_type (MessageType): The type of message to return.
        **kwargs (Any): Additional keyword arguments.

    Returns:
        Message: The output of the function action.
    """
    inputs = inputs or {} 
    return super().__call__(action_name=self.main_action_name, action_input_data=inputs, return_msg_type=return_msg_type, **kwargs)

AgentManager ¶

AgentManager(**kwargs)

Bases: BaseModule

Responsible for creating and managing all Agent objects required for workflow operation.

Attributes:

Name	Type	Description
`storage_handler`	`StorageHandler`	Used to load and save agents from/to storage.
`agents`	`List[Agent]`	A list to keep track of all managed Agent instances.
`agent_states`	`Dict[str, AgentState]`	A dictionary to track the state of each Agent by name.

Source code in evoagentx/core/module.py

def __init__(self, **kwargs):
    """
    Initializes a BaseModule instance.

    Args:
        **kwargs (Any): Keyword arguments used to initialize the instance

    Raises:
        ValidationError: When parameter validation fails
        Exception: When other errors occur during initialization
    """

    try:
        for field_name, _ in type(self).model_fields.items():
            field_value = kwargs.get(field_name, None)
            if field_value:
                kwargs[field_name] = self._process_data(field_value)
            # if field_value and isinstance(field_value, dict) and "class_name" in field_value:
            #     class_name = field_value.get("class_name")
            #     sub_cls = MODULE_REGISTRY.get_module(cls_name=class_name)
            #     kwargs[field_name] = sub_cls._create_instance(field_value)
        super().__init__(**kwargs) 
        self.init_module()
    except (ValidationError, Exception) as e:
        exception_handler = callback_manager.get_callback("exception_buffer")
        if exception_handler is None:
            error_message = get_base_module_init_error_message(
                cls=self.__class__, 
                data=kwargs, 
                errors=e
            )
            logger.error(error_message)
            raise
        else:
            exception_handler.add(e)

size `property` ¶

size

Get the total number of agents managed by this manager.

check_agents ¶

check_agents()

Validate agent list integrity and state consistency.

Performs thorough validation of the agent manager's internal state: 1. Checks for duplicate agent names 2. Verifies that agent states exist for all agents 3. Ensures agent list and state dictionary sizes match

Source code in evoagentx/agents/agent_manager.py

def check_agents(self):
    """Validate agent list integrity and state consistency.

    Performs thorough validation of the agent manager's internal state:
    1. Checks for duplicate agent names
    2. Verifies that agent states exist for all agents
    3. Ensures agent list and state dictionary sizes match
    """
    # check that the names of self.agents should be unique
    duplicate_agent_names = self.find_duplicate_agents(self.agents)
    if duplicate_agent_names:
        raise ValueError(f"The agents should be unique. Found duplicate agent names: {duplicate_agent_names}!")
    # check agent states
    if len(self.agents) != len(self.agent_states):
        raise ValueError(f"The lengths of self.agents ({len(self.agents)}) and self.agent_states ({len(self.agent_states)}) are different!")
    missing_agents = self.find_missing_agent_states()
    if missing_agents:
        raise ValueError(f"The following agents' states were not found: {missing_agents}")

has_agent ¶

has_agent(agent_name: str) -> bool

Check if an agent with the given name exists in the manager.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The name of the agent to check	required

Returns:

Type	Description
`bool`	True if an agent with the given name exists, False otherwise

Source code in evoagentx/agents/agent_manager.py

def has_agent(self, agent_name: str) -> bool:
    """Check if an agent with the given name exists in the manager.

    Args:
        agent_name: The name of the agent to check

    Returns:
        True if an agent with the given name exists, False otherwise
    """
    all_agent_names = self.list_agents()
    return agent_name in all_agent_names

load_agent ¶

load_agent(agent_name: str, tools: Optional[List[Union[Tool, Toolkit]]] = None, **kwargs) -> Agent

Load an agent from local storage through storage_handler.

Retrieves agent data from storage and creates an Agent instance.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The name of the agent to load	required
`tools`	`Optional[List[Union[Tool, Toolkit]]]`	Optional list of tools available for the agent	`None`
`**kwargs`	`Any`	Additional parameters for agent creation	`{}`

Returns:

Type	Description
`Agent`	Agent instance with data loaded from storage

Source code in evoagentx/agents/agent_manager.py

def load_agent(self, agent_name: str, tools: Optional[List[Union[Tool, Toolkit]]] = None, **kwargs) -> Agent:
    """Load an agent from local storage through storage_handler.

    Retrieves agent data from storage and creates an Agent instance.

    Args:
        agent_name: The name of the agent to load
        tools: Optional list of tools available for the agent
        **kwargs (Any): Additional parameters for agent creation

    Returns:
        Agent instance with data loaded from storage
    """
    if not self.storage_handler:
        raise ValueError("must provide ``self.storage_handler`` to use ``load_agent``")
    agent_data = self.storage_handler.load_agent(agent_name=agent_name)
    agent: Agent = create_agent_from_dict(agent_dict=agent_data, tools=tools, **kwargs)
    return agent

load_all_agents ¶

load_all_agents(**kwargs)

Load all agents from storage and add them to the manager.

Retrieves all available agents from storage and adds them to the managed agents collection.

Parameters:

Name	Type	Description	Default
`**kwargs`	`Any`	Additional parameters passed to storage handler	`{}`

Source code in evoagentx/agents/agent_manager.py

def load_all_agents(self, **kwargs):
    """Load all agents from storage and add them to the manager.

    Retrieves all available agents from storage and adds them to the
    managed agents collection.

    Args:
        **kwargs (Any): Additional parameters passed to storage handler
    """
    pass 

get_agent_name ¶

get_agent_name(agent: Union[str, dict, Agent]) -> str

Extract agent name from different agent representations.

Handles different ways to specify an agent (string name, dictionary, or Agent instance) and extracts the agent name.

Parameters:

Name	Type	Description	Default
`agent`	`Union[str, dict, Agent]`	Agent specified as a string name, dictionary with 'name' key, or Agent instance	required

Returns:

Type	Description
`str`	The extracted agent name as a string

Source code in evoagentx/agents/agent_manager.py

def get_agent_name(self, agent: Union[str, dict, Agent]) -> str:
    """Extract agent name from different agent representations.

    Handles different ways to specify an agent (string name, dictionary, or
    Agent instance) and extracts the agent name.

    Args:
        agent: Agent specified as a string name, dictionary with 'name' key,
              or Agent instance

    Returns:
        The extracted agent name as a string
    """
    if isinstance(agent, str):
        agent_name = agent
    elif isinstance(agent, dict):
        agent_name = agent["name"]
    elif isinstance(agent, Agent):
        agent_name = agent.name
    else:
        raise ValueError(f"{type(agent)} is not a supported type for ``get_agent_name``. Supported types: [str, dict, Agent].")
    return agent_name

add_agent ¶

add_agent(agent: Union[str, dict, Agent], llm_config: Optional[LLMConfig] = None, tools: Optional[List[Union[Tool, Toolkit]]] = None, **kwargs)

add a single agent, ignore if the agent already exists (judged by the name of an agent).

Parameters:

Name	Type	Description	Default
`agent`	`Union[str, dict, Agent]`	The agent to be added, specified as: - String: Agent name to load from storage - Dictionary: Agent specification to create a CustomizeAgent - Agent: Existing Agent instance to add directly	required
`llm_config`	`Optional[LLMConfig]`	The LLM configuration to be used for the agent. Only used when the `agent` is a dictionary, used to create a CustomizeAgent.	`None`
`tools`	`Optional[List[Union[Tool, Toolkit]]]`	Optional list of tools available for the agent, used to resolve tool_names in agent config.	`None`
`**kwargs`	`Any`	Additional parameters for agent creation	`{}`

Source code in evoagentx/agents/agent_manager.py

@atomic_method
def add_agent(self, agent: Union[str, dict, Agent], llm_config: Optional[LLMConfig]=None, tools: Optional[List[Union[Tool, Toolkit]]]=None, **kwargs):
    """
    add a single agent, ignore if the agent already exists (judged by the name of an agent).

    Args:
        agent: The agent to be added, specified as:
            - String: Agent name to load from storage
            - Dictionary: Agent specification to create a CustomizeAgent
            - Agent: Existing Agent instance to add directly
        llm_config (Optional[LLMConfig]): The LLM configuration to be used for the agent. Only used when the `agent` is a dictionary, used to create a CustomizeAgent.
        tools: Optional list of tools available for the agent, used to resolve tool_names in agent config.
        **kwargs (Any): Additional parameters for agent creation
    """
    agent_name = self.get_agent_name(agent=agent)
    if self.has_agent(agent_name=agent_name):
        return
    agent_instance = self.create_agent(agent=agent, llm_config=llm_config, tools=tools, **kwargs)
    self.agents.append(agent_instance)
    self.agent_states[agent_instance.name] = AgentState.AVAILABLE
    if agent_instance.name not in self._state_conditions:
        self._state_conditions[agent_instance.name] = threading.Condition()
    self.check_agents()

add_agents ¶

add_agents(agents: List[Union[str, dict, Agent]], llm_config: Optional[LLMConfig] = None, tools: Optional[List[Union[Tool, Toolkit]]] = None, **kwargs)

add several agents by using self.add_agent().

Source code in evoagentx/agents/agent_manager.py

def add_agents(self, agents: List[Union[str, dict, Agent]], llm_config: Optional[LLMConfig]=None, tools: Optional[List[Union[Tool, Toolkit]]]=None, **kwargs):
    """
    add several agents by using self.add_agent().
    """
    for agent in agents:
        self.add_agent(agent=agent, llm_config=llm_config, tools=tools, **kwargs)

add_agents_from_workflow ¶

add_agents_from_workflow(workflow_graph, llm_config: Optional[LLMConfig] = None, tools: Optional[List[Union[Tool, Toolkit]]] = None, **kwargs)

Initialize agents from the nodes of a given WorkFlowGraph and add these agents to self.agents.

Parameters:

Name	Type	Description	Default
`workflow_graph`	`WorkFlowGraph`	The workflow graph containing nodes with agents information.	required
`llm_config`	`Optional[LLMConfig]`	The LLM configuration to be used for the agents.	`None`
`tools`	`Optional[List[Union[Tool, Toolkit]]]`	Optional list of tools available for the agents, used to resolve tool_names in agent config.	`None`
`**kwargs`	`Any`	Additional parameters passed to add_agent	`{}`

Source code in evoagentx/agents/agent_manager.py

def add_agents_from_workflow(self, workflow_graph, llm_config: Optional[LLMConfig]=None, tools: Optional[List[Union[Tool, Toolkit]]]=None, **kwargs):
    """
    Initialize agents from the nodes of a given WorkFlowGraph and add these agents to self.agents.

    Args:
        workflow_graph (WorkFlowGraph): The workflow graph containing nodes with agents information.
        llm_config (Optional[LLMConfig]): The LLM configuration to be used for the agents.
        tools: Optional list of tools available for the agents, used to resolve tool_names in agent config.
        **kwargs (Any): Additional parameters passed to add_agent
    """
    from ..workflow.workflow_graph import WorkFlowGraph
    if not isinstance(workflow_graph, WorkFlowGraph):
        raise TypeError("workflow_graph must be an instance of WorkFlowGraph")
    for node in workflow_graph.nodes:
        if node.agents:
            for agent in node.agents:
                self.add_agent(agent=agent, llm_config=llm_config, tools=tools, **kwargs)

update_agents_from_workflow ¶

update_agents_from_workflow(workflow_graph, llm_config: Optional[LLMConfig] = None, tools: Optional[List[Union[Tool, Toolkit]]] = None, **kwargs)

Update agents from a given WorkFlowGraph.

Parameters:

Name	Type	Description	Default
`workflow_graph`	`WorkFlowGraph`	The workflow graph containing nodes with agents information.	required
`llm_config`	`Optional[LLMConfig]`	The LLM configuration to be used for the agents.	`None`
`tools`	`Optional[List[Union[Tool, Toolkit]]]`	Optional list of tools available for the agents, used to resolve tool_names in agent config.	`None`
`**kwargs`		Additional parameters passed to update_agent	`{}`

Source code in evoagentx/agents/agent_manager.py

def update_agents_from_workflow(self, workflow_graph, llm_config: Optional[LLMConfig]=None, tools: Optional[List[Union[Tool, Toolkit]]]=None, **kwargs):
    """
    Update agents from a given WorkFlowGraph.

    Args:
        workflow_graph (WorkFlowGraph): The workflow graph containing nodes with agents information.
        llm_config (Optional[LLMConfig]): The LLM configuration to be used for the agents.
        tools: Optional list of tools available for the agents, used to resolve tool_names in agent config.
        **kwargs: Additional parameters passed to update_agent
    """
    from ..workflow.workflow_graph import WorkFlowGraph
    if not isinstance(workflow_graph, WorkFlowGraph):
        raise TypeError("workflow_graph must be an instance of WorkFlowGraph")
    for node in workflow_graph.nodes:
        if node.agents:
            for agent in node.agents:
                agent_name = self.get_agent_name(agent=agent)
                if self.has_agent(agent_name=agent_name):
                    # use the llm_config of the existing agent
                    agent_llm_config = self.get_agent(agent_name).llm_config
                    self.update_agent(agent=agent, llm_config=agent_llm_config, tools=tools, **kwargs)
                else:
                    self.add_agent(agent=agent, llm_config=llm_config, tools=tools, **kwargs)

get_agent ¶

get_agent(agent_name: str, **kwargs) -> Agent

Retrieve an agent by its name from managed agents.

Searches the list of managed agents for an agent with the specified name.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The name of the agent to retrieve	required
`**kwargs`	`Any`	Additional parameters (unused)	`{}`

Returns:

Type	Description
`Agent`	The Agent instance with the specified name

Source code in evoagentx/agents/agent_manager.py

def get_agent(self, agent_name: str, **kwargs) -> Agent:
    """Retrieve an agent by its name from managed agents.

    Searches the list of managed agents for an agent with the specified name.

    Args:
        agent_name: The name of the agent to retrieve
        **kwargs (Any): Additional parameters (unused)

    Returns:
        The Agent instance with the specified name
    """
    for agent in self.agents:
        if agent.name == agent_name:
            return agent
    raise ValueError(f"Agent ``{agent_name}`` does not exists!")

update_agent ¶

update_agent(agent: Union[dict, Agent], llm_config: Optional[LLMConfig] = None, tools: Optional[List[Union[Tool, Toolkit]]] = None, **kwargs)

Update an agent in the manager.

Parameters:

Name	Type	Description	Default
`agent`	`Union[dict, Agent]`	The agent to be updated, specified as: - Dictionary: Agent specification to update a CustomizeAgent - Agent: Existing Agent instance to update	required
`llm_config`	`Optional[LLMConfig]`	The LLM configuration to be used for the agent.	`None`
`tools`	`Optional[List[Union[Tool, Toolkit]]]`	Optional list of tools available for the agent, used to resolve tool_names in agent config.	`None`

Source code in evoagentx/agents/agent_manager.py

def update_agent(self, agent: Union[dict, Agent], llm_config: Optional[LLMConfig]=None, tools: Optional[List[Union[Tool, Toolkit]]]=None, **kwargs):
    """
    Update an agent in the manager.

    Args:
        agent: The agent to be updated, specified as:
            - Dictionary: Agent specification to update a CustomizeAgent
            - Agent: Existing Agent instance to update
        llm_config (Optional[LLMConfig]): The LLM configuration to be used for the agent.
        tools: Optional list of tools available for the agent, used to resolve tool_names in agent config.
    """
    agent_name = self.get_agent_name(agent=agent)
    self.remove_agent(agent_name=agent_name)
    self.add_agent(agent=agent, llm_config=llm_config, tools=tools, **kwargs)

remove_agent ¶

remove_agent(agent_name: str, remove_from_storage: bool = False, **kwargs)

Remove an agent from the manager and optionally from storage.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The name of the agent to remove	required
`remove_from_storage`	`bool`	If True, also remove the agent from storage	`False`
`**kwargs`	`Any`	Additional parameters passed to storage_handler.remove_agent	`{}`

Source code in evoagentx/agents/agent_manager.py

@atomic_method
def remove_agent(self, agent_name: str, remove_from_storage: bool=False, **kwargs):
    """
    Remove an agent from the manager and optionally from storage.

    Args:
        agent_name: The name of the agent to remove
        remove_from_storage: If True, also remove the agent from storage
        **kwargs (Any): Additional parameters passed to storage_handler.remove_agent
    """
    self.agents = [agent for agent in self.agents if agent.name != agent_name]
    self.agent_states.pop(agent_name, None)
    self._state_conditions.pop(agent_name, None) 
    if remove_from_storage:
        self.storage_handler.remove_agent(agent_name=agent_name, **kwargs)
    self.check_agents()

get_agent_state ¶

get_agent_state(agent_name: str) -> AgentState

Get the state of a specific agent by its name.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The name of the agent.	required

Returns:

Name	Type	Description
`AgentState`	`AgentState`	The current state of the agent.

Source code in evoagentx/agents/agent_manager.py

def get_agent_state(self, agent_name: str) -> AgentState:
    """
    Get the state of a specific agent by its name.

    Args:
        agent_name: The name of the agent.

    Returns:
        AgentState: The current state of the agent.
    """
    return self.agent_states[agent_name]

set_agent_state ¶

set_agent_state(agent_name: str, new_state: AgentState) -> bool

Changes an agent's state and notifies any threads waiting on that agent's state. Thread-safe operation for coordinating multi-threaded agent execution.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The name of the agent	required
`new_state`	`AgentState`	The new state to set	required

Returns:

Type	Description
`bool`	True if the state was updated successfully, False otherwise

Source code in evoagentx/agents/agent_manager.py

@atomic_method
def set_agent_state(self, agent_name: str, new_state: AgentState) -> bool:
    """
    Changes an agent's state and notifies any threads waiting on that agent's state.
    Thread-safe operation for coordinating multi-threaded agent execution.

    Args:
        agent_name: The name of the agent
        new_state: The new state to set

    Returns:
        True if the state was updated successfully, False otherwise
    """

    # if agent_name in self.agent_states and isinstance(new_state, AgentState):
    #     # self.agent_states[agent_name] = new_state
    #     with self._state_conditions[agent_name]:
    #         self.agent_states[agent_name] = new_state
    #         self._state_conditions[agent_name].notify_all()
    #     self.check_agents()
    #     return True
    # else:
    #     return False
    if agent_name in self.agent_states and isinstance(new_state, AgentState):
        if agent_name not in self._state_conditions:
            self._state_conditions[agent_name] = threading.Condition()
        with self._state_conditions[agent_name]:
            self.agent_states[agent_name] = new_state
            self._state_conditions[agent_name].notify_all()
        return True
    return False

get_all_agent_states ¶

get_all_agent_states() -> Dict[str, AgentState]

Get the states of all managed agents.

Returns:

Type	Description
`Dict[str, AgentState]`	Dict[str, AgentState]: A dictionary mapping agent names to their states.

Source code in evoagentx/agents/agent_manager.py

def get_all_agent_states(self) -> Dict[str, AgentState]:
    """Get the states of all managed agents.

    Returns:
        Dict[str, AgentState]: A dictionary mapping agent names to their states.
    """
    return self.agent_states

save_all_agents ¶

save_all_agents(**kwargs)

Save all managed agents to persistent storage.

Parameters:

Name	Type	Description	Default
`**kwargs`	`Any`	Additional parameters passed to the storage handler	`{}`

Source code in evoagentx/agents/agent_manager.py

@atomic_method
def save_all_agents(self, **kwargs):
    """Save all managed agents to persistent storage.

    Args:
        **kwargs (Any): Additional parameters passed to the storage handler
    """
    pass 

clear_agents ¶

clear_agents()

Remove all agents from the manager.

Source code in evoagentx/agents/agent_manager.py

@atomic_method
def clear_agents(self):
    """
    Remove all agents from the manager.
    """
    self.agents = [] 
    self.agent_states = {}
    self._state_conditions = {}
    self.check_agents()

wait_for_agent_available ¶

wait_for_agent_available(agent_name: str, timeout: Optional[float] = None) -> bool

Wait for an agent to be available.

Parameters:

Name	Type	Description	Default
`agent_name`	`str`	The name of the agent to wait for	required
`timeout`	`Optional[float]`	Maximum time to wait in seconds, or None to wait indefinitely	`None`

Returns:

Type	Description
`bool`	True if the agent became available, False if timed out

Source code in evoagentx/agents/agent_manager.py

def wait_for_agent_available(self, agent_name: str, timeout: Optional[float] = None) -> bool:
    """Wait for an agent to be available.

    Args:
        agent_name: The name of the agent to wait for
        timeout: Maximum time to wait in seconds, or None to wait indefinitely

    Returns:
        True if the agent became available, False if timed out
    """
    if agent_name not in self._state_conditions:
        self._state_conditions[agent_name] = threading.Condition()
    condition = self._state_conditions[agent_name]

    with condition:
        return condition.wait_for(
            lambda: self.agent_states.get(agent_name) == AgentState.AVAILABLE,
            timeout=timeout
        )

copy ¶

copy() -> AgentManager

Create a shallow copy of the AgentManager.

Source code in evoagentx/agents/agent_manager.py

def copy(self) -> "AgentManager":
    """
    Create a shallow copy of the AgentManager.
    """
    return AgentManager(agents=self.agents, storage_handler=self.storage_handler)

Name	Type	Description
`name`	`str`	Unique identifier for the agent
`description`	`str`	Human-readable description of the agent's purpose
`llm_config`	`Optional[LLMConfig]`	Configuration for the language model. If provided, a new LLM instance will be created. Otherwise, the existing LLM instance specified in the `llm` field will be used.
`llm`	`Optional[BaseLLM]`	Language model instance. If provided, the existing LLM instance will be used.
`agent_id`	`Optional[str]`	Unique ID for the agent, auto-generated if not provided
`system_prompt`	`Optional[str]`	System prompt for the Agent.
`actions`	`List[Action]`	List of available actions
`n`	`Optional[int]`	Number of latest messages used to provide context for action execution. It uses all the messages in short term memory by default.
`is_human`	`bool`	Whether this agent represents a human user
`version`	`int`	Version number of the agent, default is 0.

Name	Type	Description	Default
`action_name`	`str`	The name of the action to execute	required
`msgs`	`Optional[List[Message]]`	Optional list of messages providing context for the action	`None`
`action_input_data`	`Optional[dict]`	Optional pre-extracted input data for the action	`None`
`return_msg_type`	`Optional[MessageType]`	Message type for the return message	`UNKNOWN`
`**kwargs`	`Any`	Additional parameters, may include workflow information	`{}`

Name	Type	Description	Default
`action_name`	`str`	Name of the action whose prompt should be updated	required
`prompt`	`str`	New prompt text to set for the action	required
`system_prompt`	`Optional[str]`	Optional new system prompt to set for the agent	`None`

Name	Type	Description	Default
`path`	`str`	Path where the agent should be saved	required
`ignore`	`List[str]`	List of field names to exclude from serialization	`[]`
`**kwargs`	`Any`	Additional parameters for the save operation	`{}`

Name	Type	Description	Default
`data`	`Dict`	The dictionary containing all necessary configuration to recreate this agent	required
`llm_config`	`Optional[LLMConfig]`	The LLMConfig instance	`None`