Creo-1

An open-source framework for developing AI Agent Networks

Enough RAG, time for MOP!

Message Orchestration Pattern

This repo implements an improvement on the RAG infrastructure, the Message Orchestration Pattern, or MOP.

This architecture aims to resolve some of the biggest problems for growing and scaling AI Agent systems:

• Efficient agent-to-agent communication
• Ease of adding new independent agents
• Abstractions for integrating with local or cloud based LLM providers
• Abstractions for custom tool calling

We accomplish this operation by implementing a robust message exchange system using a queue service provider. In our examples we will use RabbitMQ, but this is an abstract layer that can be replaced with another provider like Kafka or SQS.

Scalable Agents

By using a queue, we can create independant and scalable queue consumer agent, wich we can either pool or distribute acrosss compute resources. This allows us to scale each agent role independently.

Tech Stack

• Docker
• RabbitMQ
• MongoDB
• OpenAI (swapable service provider)

(Optional)

• Discord - used in example run\discord

Setup

Prerequisites

• Docker - (https://docs.docker.com/engine/install/)

Environment Variables

1. Copy .env.example to .env
2. Update the <values> as indicated in the file

Installation

Run the following to set up a venv and install dependencies:

make setup

Start Docker

Follow instructions for your OS to start the Docker app

Start RabbitMQ

# RabbitMQ 4.0.x
docker run -it --rm --name rabbitmq -p 5672:5672 -p 15672:15672 rabbitmq:4.0-management

Start MongoDB

export MONGODB_VERSION=6.0-ubi8
docker run --name mongodb -d mongodb/mongodb-community-server:$MONGODB_VERSION

Running examples

Simple Discord bot

Discord is an excellent tool for testing chat bots because it handles the UI overhead, allowing you to focus on feature design.

The following command will launch a simple discord bot using the example located at run/local/discord_bot.py.

make run-local

You will need to make sure you have the following vaues defined in your .env file:

DISCORD_CHANNEL_ID=<your-channel-id>
DISCORD_TOKEN=<your-discord-token>