Agent

The Agent class is the base class for all agents. It provides the main logic for running the agent and processing the responses from the LLM.

It is an abstract class that should be inherited by all agents. The class provides the main logic for running the agent and processing the responses from the LLM.

GenericAgent class implements all abstract functions for generic version of an agent in Elemental.

class Agent(ABC):
    """
    Generic Agent class.
    """

    def __init__(
        self,
        agent_logic: AgentLogic,
        short_memory_capacity: int = -1,
        toolbox: ToolBox = None,
        termination_sequence: str = None,
    ) -> None

AgentLogic object brings all information related to language model interaction including prompt, template and all variables used in the prompt (name, persona).

The Agent class includes the following methods:

run - Run the agent's main logic function in the iterative fashion.

@abstractmethod
def run(
        self, 
        task: str | List[str], 
        input_session: str
    ) -> str:

run_instruction - Run the agent's main logic function on a single instruction. Tuple of a boolean indicating if the response is terminal and the string response.

@abstractmethod
def run_instruction(
        self, 
        instruction: str, 
        original_instruction: str = "", 
        input_session: str = ""
    ) -> Tuple[bool, str]

run_instruction_inference - Run the agent's main logic function on a single instruction inference, only LLM portion of agent logic.

@abstractmethod
def run_instruction_inference(
        self, 
        instruction: str, 
        original_instruction: str = "", 
        input_session: str = ""
    ) -> str

run_instruction_action - Run the agent's requested actions. Execute tools and return the result.

@abstractmethod
def run_instruction_action(
        self, 
        agent_response: str
    ) -> str

process_response - Process the response from the LLM. This method should be used to post-process the raw response and extract only the final result.
```
@abstractmethod
def process_response(
        self, 
        response: str
    ) -> str
```
get_termination_sequence - Get the termination sequence for the agent.
```
def get_termination_sequence(self) -> str
```
get_agent_name - Get the agent's name.
```
def get_agent_name(self) -> str
```
get_agent_persona - Get the agent's persona.
```
def get_agent_persona(self) -> str
```
reset_short_memory - Reset the agent's short memory. This method should be used to clear the short memory of the agent.
```
def reset_short_memory(self) -> None
```
get_all_messages - Get all messages from the agent's short memory.
```
def get_all_messages(self) -> List[Message]
```

Instruction

Agent Team