LLM Tools Architecture¶

Overview¶

The PigPig Bot's LLM can interact with the world through a set of specialized Tools. These tools allow the AI to search the internet, manage reminders, perform complex math, and access the bot's internal memory systems.

The Tool Layer¶

All tools are located in the llm/tools/ directory. Each tool module typically exposes a *Tools class that acts as a container for one or more LangChain-compatible tools.

`ToolsFactory`¶

The ToolsFactory (in llm/tools_factory.py) is responsible for: 1. Discovery: Dynamically finding tool classes in the llm/tools/ package. 2. Instantiation: Injecting the RuntimeContext (bot instance, channel info, etc.) into the tools. 3. Filtering: Enabling or disabling tools based on the bot's configuration and guild permissions.

Anatomy of a Tool¶

A typical tool module follows this pattern:

class MyTools:
    def __init__(self, runtime):
        self.runtime = runtime

    def get_tools(self):
        @tool
        async def my_custom_tool(param: str) -> str:
            """Tool description used by LLM."""
            # Implementation
            return "result"
        return [my_custom_tool]

Available Tool Categories¶

Information Retrieval: Internet search, YouTube info, Wikipedia.
Internal Knowledge: User stats, server context, memory retrieval.
Utility: Math operations, reminders, image generation.
Meta-Tools: list_tools for discovering other available tools.

For a complete reference of available tools and their signatures, see the Tool List.