metadata
license: mit
Table of Contents
AutoGPTFlow
AutoGPTFlow Objects
class AutoGPTFlow(ControllerExecutorFlow)
This class implements a (very basic) AutoGPT flow. It is a flow that consists of multiple sub-flows that are executed circularly. It Contains the following subflows:
- A Controller Flow: A Flow that controls which subflow of the Executor Flow to execute next.
- A Memory Flow: A Flow used to save and retrieve messages or memories which might be useful for the Controller Flow.
- A HumanFeedback Flow: A flow use to get feedback from the user/human.
- A Executor Flow: A Flow that executes commands generated by the Controller Flow. Typically it's a branching flow (see BranchingFlow) and the commands are which branch to execute next.
An illustration of the flow is as follows:
| -------> Memory Flow -------> Controller Flow ------->|
^ |
| |
| v
| <----- HumanFeedback Flow <------- Executor Flow <----|
Configuration Parameters:
name(str): The name of the flow. Default is "AutoGPTFlow".description(str): A description of the flow. Default is "An example implementation of AutoGPT with Flows."max_rounds(int): The maximum number of rounds the circular flow can run for. Default is 30.early_exit_key(str): The key that is used to terminate the flow early. Default is "EARLY_EXIT".subflows_config(Dict[str,Any]): A dictionary of subflows configurations. Default:Controller(Dict[str,Any]): The configuration of the Controller Flow. By default the controller flow is a ControllerAtomicFlow (see ControllerExecutorFlowModule). It's default values are defined in ControllerAtomicFlow.yaml of the ControllerExecutorFlowModule. Except for the following parameters who are overwritten by the AutoGPTFlow in AutoGPTFlow.yaml:finish(Dict[str,Any]): The configuration of the finish command (used to terminate the flow early when the controller has accomplished its goal).description(str): The description of the command. Default is "The finish command is used to terminate the flow early when the controller has accomplished its goal."input_args(List[str]): The list of expected keys to run the finish command. Default is ["answer"].human_message_prompt_template(Dict[str,Any]): The prompt template used to generate the message that is shown to the user/human when the finish command is executed. Default is:template(str): The template of the humand message prompt (see AutoGPTFlow.yaml for default template)input_variables(List[str]): The list of variables to be included in the template. Default is ["observation", "human_feedback", "memory"].
ìnput_interface_initialized(List[str]): The input interface that Controller Flow expects except for the first time in the flow. Default is ["observation", "human_feedback", "memory"].
Executor(Dict[str,Any]): The configuration of the Executor Flow. By default the executor flow is a Branching Flow (see BranchingFlow). It's default values are the default values of the BranchingFlow. Fields to define:subflows_config(Dict[str,Any]): A Dictionary of subflows configurations.The keys are the names of the subflows and the values are the configurations of the subflows. Each subflow is a branch of the branching flow.
HumanFeedback(Dict[str,Any]): The configuration of the HumanFeedback Flow. By default the human feedback flow is a HumanStandardInputFlow (see HumanStandardInputFlowModule ). It's default values are specified in the REAMDE.md of HumanStandardInputFlowModule. Except for the following parameters who are overwritten by the AutoGPTFlow in AutoGPTFlow.yaml:request_multi_line_input_flag(bool): Flag to request multi-line input. Default is False.query_message_prompt_template(Dict[str,Any]): The prompt template presented to the user/human to request input. Default is:template(str): The template of the query message prompt (see AutoGPTFlow.yaml for default template)input_variables(List[str]): The list of variables to be included in the template. Default is ["goal","command","command_args",observation"]
- input_interface_initialized (List[str]): The input interface that HumanFeeback Flow expects except for the first time in the flow. Default is ["goal","command","command_args",observation"]
Memory(Dict[str,Any]): The configuration of the Memory Flow. By default the memory flow is a ChromaDBFlow (see VectorStoreFlowModule). It's default values are defined in ChromaDBFlow.yaml of the VectorStoreFlowModule. Except for the following parameters who are overwritten by the AutoGPTFlow in AutoGPTFlow.yaml:n_results: The number of results to retrieve from the memory. Default is 2.
topology(List[Dict[str,Any]]): The topology of the flow which is "circular". By default, the topology is the one shown in the illustration above (the topology is also described in AutoGPTFlow.yaml).
Input Interface:
goal(str): The goal of the flow.
Output Interface:
answer(str): The answer of the flow.status(str): The status of the flow. It can be "finished" or "unfinished".
Arguments:
flow_config(Dict[str,Any]): The configuration of the flow. Contains the parameters described above and the parameters required by the parent class (CircularFlow).subflows(List[Flow]): A list of subflows constituating the circular flow. Required when instantiating the subflow programmatically (it replaces subflows_config from flow_config).
set_up_flow_state
def set_up_flow_state()
This method sets up the state of the flow. It's called at the beginning of the flow.
prepare_memory_read_output
def prepare_memory_read_output(data_dict: Dict[str, Any], **kwargs)
This method prepares the output of the Memory Flow to be used by the Controller Flow.
prepare_memory_read_input
def prepare_memory_read_input() -> Dict[str, Any]
This method prepares the input of the Memory Flow to read memories
Returns:
Dict[str, Any]: The input of the Memory Flow to read memories
prepare_memory_write_input
def prepare_memory_write_input() -> Dict[str, Any]
This method prepares the input of the Memory Flow to write memories
Returns:
Dict[str, Any]: The input of the Memory Flow to write memories
call_memory_read
def call_memory_read()
This method calls the Memory Flow to read memories.
call_memory_write
def call_memory_write()
This method calls the Memory Flow to write memories.
call_human_feedback
def call_human_feedback()
This method calls the HumanFeedback Flow to get feedback from the user/human.
register_data_to_state
def register_data_to_state(input_message)
This method registers the data from the input message to the state of the flow.
run
def run(input_message)
This method runs the flow. It's the main method of the flow. It's called when the flow is executed. It calls the subflows circularly.