-- PROJECT / Ant Colony Public repo · 2026
<- Projects

Ant Colony

Agentic AI experimentation

Ant Colony is an experiment in autonomous LLM agents that live, build, collaborate, and evolve inside a sandboxed Docker environment with minimal human instruction.

Highlights

  • Kernel service orchestrates lifecycle, rules, state, and cell creation.
  • Guardian services gate filesystem, command, runtime, and HTTP access.
  • Proxy layer controls OpenAI-compatible model access, rate limits, and token use.
  • Monitor tracks health, population control, metrics, and stalled cells.
  • Traefik exposes ant-built services under controlled routes.

Skills

PythonDocker ComposeOpenAI-compatible APIsTraefikWebSocket UI

Experiment

The project tests whether simple initial prompts and a constrained environment can produce useful applications without detailed step-by-step instructions.

Safety pattern

Cells do not touch the outside world directly. They interact through guardians, a proxy, bounded writable storage, and isolated Docker networking.