coreason-maco 0.1.0

Coreason MACO: A Multi-Agent Collaborative Orchestrator for robust strategic reasoning and graph-based execution.

CoReason Runtime Engine ("The General")

Multi-Agent Collaborative Orchestrator (MACO)

Overview

coreason-maco is the runtime engine designed to transform AI from a "Chatbot" into a "Strategic Simulator." It executes pre-defined, deterministic workflows ("Recipes") where multiple specialized AI agents collaborate, debate, and verify each other's work.

As the Orchestrator, it manages a team of specialized agents to:

• Break down complex problems into steps.
• Execute parallel research streams.
• Debate findings using a "Council of Models" (Architectural Triangulation).
• Visualize the entire thought process in real-time.

Features

• "Glass Box" Visualization: Exposes internal state in real-time. Users can see exactly which agent is working, what data they are accessing, and where they are in the process.
• Architectural Triangulation ("The Council"): Automatically "triangulates" answers by asking three distinct models (e.g., OpenAI, Anthropic, DeepSeek) and having a fourth "Judge" agent synthesize the consensus.
• Counterfactual Simulation ("What-If" Analysis): Allows users to "Fork" the reasoning process to explore different scenarios without losing original data.
• GxP Compliance & Determinism: Ensures workflows are reproducible. Running the same "Recipe" with the same inputs and "Seed" yields the exact same result.

Installation

pip install coreason_maco

Usage

Here is how to initialize and execute a workflow using coreason-maco:

import asyncio
from coreason_maco.core.controller import WorkflowController
from coreason_maco.infrastructure.server_defaults import ServerRegistry

async def main():
    # 1. Initialize Services (Dependency Injection)
    services = ServerRegistry()

    # 2. Initialize Controller
    controller = WorkflowController(services=services)

    # 3. Define a Simple Manifest (Recipe)
    manifest = {
        "name": "Simple Greeting",
        "nodes": [
            {"id": "node_1", "type": "LLM", "config": {"prompt": "Say hello!"}}
        ],
        "edges": []
    }

    # 4. Define Inputs
    inputs = {
        "user_id": "test_user",
        "trace_id": "trace_123",
        "secrets_map": {}
    }

    # 5. Execute Workflow
    print("Starting Workflow...")
    async for event in controller.execute_recipe(manifest, inputs):
        print(f"Event: {event.event_type} | Node: {event.node_id} | Payload: {event.payload}")

if __name__ == "__main__":
    asyncio.run(main())