Multi-agent agentic company — AI agents that plan, build, test, and ship an eSports data platform, steered by humans who each spend max 1 hour/day on decisions.
The architecture follows the Roman Colosseum model: two inseparable buildings connected by a tunnel.
The Colosseum (arena/) — the organizational framework: constitution, operating model, role definitions, and the rules that govern how agents collaborate.
The Ludus Magnus (ludus/) — where agents are made: SOULs, GPG identities, deployment configs, and the skills registry. Agents enter the arena through the crypta — session-start injection of their SOUL and ROLE.
The crypta — the tunnel between them. When SOUL.md and ROLE.yaml are loaded at session-start, the agent is no longer a definition in a file — it is an active participant in the framework.
Each human runs an independent node with their own Apex (Chief of Staff) and agent fleet. Nodes coordinate through GitHub.
Human A Human B
│ │
▼ ▼
[ Apex — CoS ] [ Apex — CoS ]
│ │
├── Product Wing ├── Product Wing
├── Engineering Wing ├── Engineering Wing
├── Quality Wing ├── Quality Wing
└── Domain Wing └── Domain Wing
│ │
└────────── GitHub ──────────────┘
Issues + PRs
(shared work repos)
13 agents across 5 wings per node: Priya (PM), Maren (UX), Atlas (Architect), Rio (EM), Vite, Forge, Muse, Hertz, Helm (Engineering), Probe (QA), Senna, Quill (Domain).
Our research into agent architectures, coordination patterns, and deployment strategies is public:
- exploration-openclaw — architecture deep-dives, agent communication patterns, and deployment research