Toolset Configuration¶

Toolsets are collections of tools that can be dynamically loaded from various sources. They provide a way to organize and manage groups of related tools.

OpenAPI Toolset¶

Creates tools from an OpenAPI/Swagger specification.

agents:
  api-agent:
    toolsets:
      - type: "openapi"  # toolset discriminator
        namespace: "github"  # optional prefix for tool names
        spec: "https://api.github.com/openapi.json"  # URL or path to spec
        base_url: "https://api.github.com"  # optional base URL override

Entry Points Toolset¶

Loads tools registered through Python entry points.

agents:
  plugin-agent:
    toolsets:
      - type: "entry_points"
        namespace: "my_plugins"  # optional namespace prefix
        module: "my_package"  # Python module containing entry points

Custom Toolset¶

Creates tools from a custom Python class implementation.

agents:
  custom-agent:
    toolsets:
      - type: "custom"
        namespace: "aws"  # optional namespace prefix
        import_path: "myapp.toolsets.AWSToolSet"  # path to toolset class

Toolset Class Example¶

Example of implementing a custom toolset:

from llmling.tools.toolsets import ToolSet
from llmling_agent.tools.base import Tool

class MyToolSet(ToolSet):
    """Custom toolset implementation."""

    def __init__(self, namespace: str | None = None):
        super().__init__(namespace=namespace)

    async def get_tools(self) -> list[Tool]:
        """Return list of tools in this set."""
        tools = []
        # Create and return tools
        return tools

Configuration Notes¶

The type field serves as discriminator for toolset types
Namespaces help prevent tool name collisions
Toolsets are loaded when the agent initializes
Custom toolsets must inherit from ToolSet
OpenAPI specs can be local files or URLs

Entry points should use the format:

[my_package.tools]
tool_name = path.to:function

Tools from toolsets can be filtered through agent capabilities
Environment variables can be referenced in configuration