Agent

The miminions.agent module is the core reasoning engine — an async agent powered by pydantic-ai and OpenRouter.

Features

• Async Execution — owns its full async run() reasoning loop
• Context Injection — attach a workspace and the agent builds a dynamic system prompt via ContextBuilder before every turn
• Tool Registration — register Python functions as LLM-callable tools
• MCP Integration — load tools directly from Model Context Protocol servers

Quick Start

import asyncio
from miminions.agent import create_minion

async def main():
    agent = create_minion("MyAgent")

    def add(a: int, b: int) -> int:
        return a + b

    agent.register_tool("add", "Add two numbers", add)
    reply = await agent.run("What is 3 + 7?")
    print(reply)

asyncio.run(main())

Attaching a Workspace

from miminions.agent import create_minion
from miminions.core.workspace import WorkspaceManager

agent = create_minion("MyAgent")

manager = WorkspaceManager()
workspace = manager.get_workspace("my_workspace_id")

agent.set_context(workspace, root_path="./my_workspace")

When set_context is called, ContextBuilder assembles all workspace facts, rules, and global memory insights into the system prompt on every LLM call.

Loading MCP Tools

agent = create_minion("MyAgent")
await agent.load_mcp_tools("path/to/mcp_server.py")

API Reference

create_minion(name: str) -> Minion

Factory function that creates a configured Minion instance.

Minion

Method	Description
async run(prompt)	Send a prompt and get a reply
register_tool(name, desc, func)	Register a Python function as a tool
set_context(workspace, root_path)	Attach a workspace for context injection
async load_mcp_tools(server_path)	Load tools from an MCP server
list_tools()	List all registered tool names