∞ Conversable Agent (Adv)

Documentation

• Class name: ConversableAgentCreatorAdvanced
• Category: SALT/Language Toolkit/Agents
• Output node: False

This node facilitates the advanced creation of conversable agents, allowing for the customization of agent attributes such as name, system message, and optional parameters like LLM model, default auto-reply, and a description. It's designed to enable the creation of sophisticated conversational AI agents that can interact with users, process document QA tasks, and provide automated responses.

Input types

Required

• name
  • The name of the conversable agent being created. It serves as a unique identifier and a way to reference the agent in interactions.
  • Comfy dtype: STRING
  • Python dtype: str
• system_message
  • A system message that defines the agent's initial instructions or guidelines, setting the context for its operation.
  • Comfy dtype: STRING
  • Python dtype: str

Optional

• llm_model
  • An optional parameter specifying the language learning model to be used by the agent, enhancing its conversational capabilities.
  • Comfy dtype: LLM_MODEL
  • Python dtype: dict
• default_auto_reply
  • The default message the agent will use when it cannot generate a response based on the LLM model or code execution.
  • Comfy dtype: STRING
  • Python dtype: str
• description
  • A brief description of the agent, which can be used for identification or informational purposes by other agents.
  • Comfy dtype: STRING
  • Python dtype: str

Output types

• agent
  • Comfy dtype: AGENT
  • The conversable agent created by this node, ready for interaction and capable of processing document QA tasks.
  • Python dtype: ConversableAgent

Usage tips

• Infra type: CPU
• Common nodes: unknown

Source code

class ConversableAgentCreatorAdvanced:
    @classmethod
    def INPUT_TYPES(cls):
        return {
            "required": {
                "name": ("STRING", {"multiline": False, "placeholder": "Assistant"}),
                "system_message": ("STRING", {
                    "multiline": True,
                    "default": "You are a helpful AI assistant. You can help with document QA. Return 'TERMINATE' when the task is done."
                }),
            },
            "optional": {
                "llm_model": ("LLM_MODEL",),
                # default auto reply when no code execution or llm-based reply is generated.
                "default_auto_reply": ("STRING", {"multiline": True}),
                # a short description of the agent, this description is used by other agents.
                "description": ("STRING", {"multiline": True}),
            }
        }

    RETURN_TYPES = ("AGENT",)
    RETURN_NAMES = ("agent",)

    FUNCTION = "create_agent"
    CATEGORY = f"{MENU_NAME}/{SUB_MENU_NAME}/Agents"

    def create_agent(self, name, system_message, llm_model=None, default_auto_reply="", description=None):
        agent = ConversableAgent(
            name=name,
            system_message=system_message,
            llm_config={"config_list": [{"model": llm_model["llm"].model, "api_key": llm_model["llm"].api_key}]} if llm_model is not None else False,
            human_input_mode="NEVER",
            default_auto_reply=default_auto_reply,
            description=description if description is not None else system_message,
        )
        return (agent,)