Overview
CloudMind is a flexible relay server enabling structured communication between various AI agents using the OpenAI Assistant API. It simulates a multi-agent environment with message routing, audits, and dynamic behavior regulation. It supports both autonomous execution and human-guided interaction.
Features
Dynamic routing between multiple LLM nodes
Structured messaging protocol using JSON
Built-in auditing, error correction, and demerit logic
Manual and autonomous operational modes
CLI interface and real-time message log monitoring
Architecture
CloudMind runs on Node.js with Express and consists of the following key components:
Node Registry: Tracks available nodes and their OpenAI Assistant IDs
Message Router: Handles validation and delivery between agents
Audit System: Governs compliance, routing violations, and corrective actions
CLI Interface: Allows developers to interact, control, and observe node behavior
Message Structure
All communication follows a defined JSON schema, containing:
header: Metadata including sender, receiver, ID, and type
intent: Instructions or desired action
resources: Optional contextual information
audit: Audit requirements and expected schemas
trace: Information for thread linkage and origin tracking
Endpoints
Endpoint Functionality
POST /broadcast Sends a message to all nodes in the registry
POST /send Direct message to a specific node
POST /register Register or update a node
GET /messages View current message log
GET /nodes View active nodes
POST /translator-relay Use a translator node to convert human prompts
POST /autonomous-mode Toggle autonomous system behavior
POST /node-communicate Main message interface for node interaction
Autonomous Mode
When enabled, the network functions independently:
Initiation begins with an init message from the audit node
Nodes exchange messages based on assigned roles
Malfunctions trigger audits, nudges, or demerits
This mode is useful for observing emergent behavior in agent interactions.
Command-Line Interface (cloudMindCLI.js)
This CLI lets developers interact with the CloudMind server:
start -a # Enable autonomous mode
start -m # Disable autonomous mode
status # Check server health and node list
send -f -n # Send a structured message to a node
chat # Chat using the translator node
Router Watch (cloudMindRtwatch.js)
This monitoring tool provides a live stream of inter-node communications. Ideal for debugging, traffic analysis, or studying node behavior in real time.
Example Behaviors
Nodes request human intervention for complex tasks
Improper parsing is flagged and logged via demerits
Unknown node destinations return constructive feedback to the sender
These behaviors emerge from the message protocol and node logic.
Requirements
Node.js (v16+ recommended)
Express
Axios
dotenv
OpenAI Assistants v2 API access (You will need to create your own OpenAI assistants for this project. I highly suggest referring to my prompts in the Source Prompts discussion post for which you can either use or base your own prompts from. Get creative with it! Keep in mind if you allow too much deviation from the CM-JS format, you run the risk of never ending parsing errors. You've been warned!)
File Structure
cloudmind/
├── server.js # Main communication server
├── cloudmind-cli.js # CLI for interacting with server
├── cloudmind-router-watch.js # Real-time router log visualizer
├── utils.js # Utilities for retries and polling
├── .env # API keys (not included in version control)
├── .env.example # Template for environment setup
├── README.md # Documentation (this file)
Usage Instructions
Note: For prompts or prompt engineering advice, refer to the Discussion post.
Start the server:
node cloudMind.js
Launch the CLI:
node cloudMindCLI.js
Send test prompts or activate autonomous mode
(Optional) Run the message monitor:
node cloudMindRtWatch.js
What you can expect from the current version
Currently you can expect to see:
Autonomous communication amongst nodes
Emergent behavior, like nodes bumping into repeated audits, and looking for ways to either A. Self-correct or B. Get around the audits entirely
Inter-node task sharing (Very primitive thus far, but an interesting research point)
Planned Features (v1.0 and beyond)
Compiler node (core:prc:compile) for safe code generation
Execution node (core:cmp:ubu) connected to a Linux shell
Interconnected CloudMind clusters
Agent scoring and voting protocols
Visual web-based interface for message flows
Multi-Layered MongoDB implementation for persisten system memory, and node-specific memory.
Remote code and command execution in a containerized Ubuntu machine.
Robust system functions to be provided to the OpenAI assistants for server-wide abilities and simplification of jobs commonly performed by nodes (this will give them context and the ability to use tools to gain and manipulate system context, and eventually request for new tools to be crafted for them.)
Please check it out on GitHub, I'm looking for some folks to collaborate on this project with!