This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine.
SOMAS: Giving Every Person Their Own AI — Ghost in the Machine
What if the next generation of artificial intelligence wasn't another chatbot?
What if every person could create their own AI, give it responsibilities, customize its capabilities, and connect it with other people's AI systems?
This is the vision behind SOMAS.
I believe the future of AI isn't only about building smarter models. It's about making intelligent agents an accessible part of everyday human life.
What I Built
I started developing SOMAS to explore a world where people can create and manage their own AI systems.
The project has two major concepts.
1. Root SOMAS — Create Your Own AI Organization
Root SOMAS allows us to explore how specialized AI agents could work together under human direction.
Imagine a small business owner creating a team of AI agents:
- A Manager that coordinates responsibilities
- An Assistant that helps with everyday activities
- A Research Agent that gathers relevant information
- A Virtual Customer that evaluates the user experience
- Specialized agents that handle particular tasks
Instead of asking one chatbot to do everything, users could create an organization of intelligent agents.
Our current software alpha includes persistent agent configurations, task orchestration, a Manager interface, structured change proposals, and owner approval for supported operations.
These are early foundations. The platform does not yet autonomously implement arbitrary software changes or operate a fully independent AI organization.
2. SOMAS Network — Connecting Personal AI Systems
The longer-term goal is to connect independently controlled SOMAS agents into a decentralized network.
Imagine a future where your personal AI can communicate with a restaurant's AI, arrange an appointment, coordinate a service, or collaborate with another person's agents.
I'm exploring blockchain technology for verifiable identity, ownership records, authorization, and trusted transactions.
Routine conversations and sensitive personal data should remain off-chain.
The SOMAS Network is still a proposed architecture, rather than a completed blockchain deployment.
How This Connects to “Ghost in the Machine”
The challenge theme made me think about an interesting question:
What happens when software stops feeling like a collection of tools and begins functioning as a coordinated set of intelligent agents?
With SOMAS, I want to explore how specialized agents can maintain distinct responsibilities, interact with one another, and contribute to something larger than any individual agent.
The “ghost” is not a claim that AI is conscious or alive. It represents the appearance of coordinated intelligence emerging from multiple software components.
For me, this is the most fascinating part of agent-based computing.
Demo
I've been developing an interactive concept demonstration to illustrate SOMAS and how its agents might operate.
Interactive demonstration: [INSERT YOUR VERIFIED SOMAS DEMO URL]
The demonstration explores agent organizations, communication, and potential network interactions.
Some interactions are conceptual or simulated. I want to be transparent about the difference between the current alpha implementation and the longer-term SOMAS vision.
Code
GitHub repository:
https://github.com/danielmorales839000-byte/somas
The repository contains the SOMAS v2 alpha implementation, including backend functionality, agent orchestration, database definitions, UI components, and testing infrastructure.
The source code should be evaluated separately from the broader conceptual network visualizations.
How I Built It
The SOMAS alpha is built using a combination of web infrastructure, agent orchestration logic, and optional external language-model integrations.
The main components include:
Backend and infrastructure
- JavaScript and Node.js tooling
- Cloudflare Workers for application execution
- Neon PostgreSQL for persistent data
Agent orchestration
- Manager and Assistant agent roles
- Virtual Customer and task-specific agents
- Separate persisted policy configurations
- Structured task proposals and explicit owner approval
- Agent feedback and controlled policy evaluation
AI integration
- Optional integrations through Groq and OpenRouter
- Local deterministic routing and orchestration
- Model-independent application interfaces
Testing
- Automated JavaScript tests
- PostgreSQL integration testing
- Playwright browser testing
AI model calls are optional in the present alpha. I have not yet established a challenge-specific demonstration of an open-weight model running the completed SOMAS workflow.
The important engineering problem I'm exploring is how independent agents can coordinate reliably while preserving hum
Top comments (0)