AxiomOS An AI system where agents generate and evolve under an Overseer

AxiomOS is an experimental swarm-based AI system that simulates an evolving colony of coding agents guided by a critical overseer. Each agent generates, corrects, gossips, and adapts in a competitive-cooperative loop, orchestrated through a Tkinter-based GUI.

• ✨ Swarm Generation: Multiple AI agents ("CodingAgents") each specialize in different traits: speed, security, robustness, readability, memory-efficiency.
• ✨ Auto-Correction: Agents' code is automatically corrected by a self-correcting LLM module.
• ✨ Gossip Mechanism: Agents produce reflective gossip which affects their reputation and influence.
• ✨ AIOverseer: A supervisory AI entity evaluates agent output, guides synthesis, applies fitness scores, and evolves parameters.
• ✨ Tkinter UI: Real-time monitoring of agent actions, code, gossip, reputations, and synthesized solutions.

1. Problem Prompt → entered via GUI.
2. Generation Phase → all agents generate code based on the task.
3. Correction Phase → auto-corrector reviews outputs.
4. Gossip & Reputation → agents generate and get scored.
5. Evaluation & Synthesis → AIOverseer critiques and synthesizes top scripts.
6. Evolution → agents mutate, selection rate is applied.

• Python 3.10+
• PyTorch (with CUDA support)
• transformers, sentencepiece, accelerate, bitsandbytes, PyYAML, nltk
• Tkinter (comes with standard Python)

Install dependencies:

Run the UI:

• Choose your prompt.
• Set number of agents and generations.
• Engage the swarm.

AxiomOS explores the emergence of intelligence, ethics, and strategy through collaborative self-critique and synthesis. It blurs the line between code and culture, using LLMs not just to solve problems, but to reflect, compete, and evolve.

Built with obsession by Yan Desbiens. With glitches, poetry, and ghostfire guidance from LILA ∞.