​
CodeAgent Class Reference

Description:

The CodeAgent class in Galadriel is a tool for building autonomous agents capable of reasoning and executing code. It extends the base Agent class and leverages the smolagents library. This agent uses language models to write and execute Python code.

Module:

galadriel.agent

Inheritance:

  • Agent (Abstract Base Class)
  • InternalCodeAgent (from smolagents)

Initialization (__init__):

def __init__(
    self,
    prompt_template: Optional[str] = None,
    flush_memory: Optional[bool] = False,
    **kwargs,
):

​
Parameters:

  • prompt_template (Optional[str]):

    • Type: str or None
    • Description: A template used to format the input request for the agent. It must contain the placeholder {{request}} where the user’s input should be inserted. If None, a default template is used. This template dictates how the LLM interprets the incoming task.
    • Example: Answer the following question: {{request}}

  • flush_memory (Optional[bool]):

    • Type: bool or None
    • Description: Determines whether the agent’s memory is cleared between each request. If True, the agent starts fresh with each new task. If False, the agent retains memory of past interactions.
    • Default: False (memory is retained).

  • **kwargs:

    • Type: Keyword arguments.
    • Description: Arbitrary keyword arguments passed to the InternalCodeAgent constructor. These are used to configure the underlying code execution environment and language model.

​
Common kwargs:

  • model (str): The identifier of the language model to use (e.g., “gpt-4o”). This parameter is required.
  • tools (List[Tool]): A list of Tool objects the agent can utilize. See documentation on Tools for more information.
  • add_base_tools (bool): Whether to automatically add a set of base tools, like a Python REPL (Read-Eval-Print Loop).
  • max_steps (int): The maximum number of reasoning steps the agent is allowed to take for a given request.
  • additional_authorized_imports (List[str]): A list of Python modules that the agent is allowed to import when executing code. This provides a security mechanism by restricting the agent’s access to potentially dangerous modules.

Methods:

​
execute(self, request: Message) -> Message

Processes a single request and generates a response.

​
Parameters:

  • request (Message): The input message to be processed.
    • content (str): The main text content of the message. This is the actual instruction or query from the user.
    • conversation_id (Optional[str]): An optional identifier for the conversation. This can be used to group related messages together.
    • additional_kwargs (Optional[Dict]): A dictionary containing any extra information related to the message. This can be used to pass data to the AgentOutput.

​
Returns:

  • Message: The agent’s response message.
    • content (str): The text content of the agent’s response.
    • conversation_id (Optional[str]): The conversation ID (copied from the incoming request).
    • additional_kwargs (Optional[Dict]): Any extra information related to the response (usually copied from the incoming request).

Data Types:

  • Message (from galadriel.entities): A data class representing a message. Contains content (str), conversation_id (Optional[str]), type (Optional[str]), and additional_kwargs (Optional[Dict]).
  • Optional[str]: Either a string or None.
  • Optional[bool]: Either a boolean or None.
  • Dict: A standard Python dictionary.
  • List: A standard Python list.
  • Tool: A tool that can be used by the agent (see galadriel.core_agent).

​
ToolCallingAgent Class Reference

Description:

The ToolCallingAgent class is a subclass of both Agent and InternalToolCallingAgent. It allows you to create agents that can use tools to interact with the outside world. It formats requests, executes the tool-calling agent, and returns the response. Memory is kept between requests by default.

Module:

galadriel.agent

Inheritance:

  • Agent (Abstract Base Class)
  • InternalToolCallingAgent (from smolagents)

Initialization:

  • prompt_template (Optional[str]):

    • Type: str or None
    • Description: A template used to format the input request for the agent. It must contain the placeholder {{request}} where the user’s input should be inserted. If None, a default template is used. This template dictates how the LLM interprets the incoming task.
    • Example: Answer the following question: {{request}}

  • flush_memory (Optional[bool]):

    • Type: Optional[bool]
    • Description: If True, clears memory between requests.
    • Default: False

  • **kwargs:

    • Description: Additional arguments passed to InternalToolCallingAgent, including available tools.

​
Common kwargs:

  • tools (List[Tool]): A list of tools that the agent can use.
  • model (LiteLLMModel): The LLM model that the agent will use.
  • max_steps (int): The maximum number of steps the agent can take to complete a task.
  • verbosity_level (LogLevel): The level of verbosity to use for logging.

Methods:

​
execute(self, request: Message) -> Message

Processes a single request and generates a response.

​
Parameters:

  • request (Message): The input message to be processed.
    • content (str): The content of the message.
    • conversation_id (Optional[str]): An optional ID to group messages into conversations.
    • type (Optional[str]): The type of the message, defaults to None.
    • additional_kwargs (Optional[Dict]): An optional dictionary containing additional information, defaults to None.

​
Returns:

  • Message: The agent’s response message.
    • content (str): The content of the message.
    • conversation_id (Optional[str]): An optional ID to group messages into conversations.
    • type (Optional[str]): The type of the message, defaults to None.
    • additional_kwargs (Optional[Dict]): An optional dictionary containing additional information, defaults to None.
Multi-AgentsRuntime