Private AI operations systems. We design, deploy and operate AI workers for inboxes, documents, recruiting, reporting and company-specific workflows — running on infrastructure you control, with an audit-proof trail of every AI decision.
What makes it different: the fleet improves itself. Agents capture their own lessons, review their workflows and propose amendments — recursive self-improvement (RSI) applied to operating a business, with humans approving every change.
- 22 AI agents in production, each with its own desktop, browser, tools and memory
- 8-node NVIDIA DGX Spark cluster + 4× RTX 3090 — all self-hosted, no data leaves our hardware
- Custom workflows run by herds of agents (we moved past n8n-style flowcharts)
- Odoo-based business systems wired to AI agents
| Repo | What it is |
|---|---|
| herdr-factory-loop-skill | m2herd — orchestrate herds of AI coding agents in a spec-driven factory loop (spec → plan → tasks → implement → review → evolve) |
| fleet-playbook | The operating model for autonomous AI organisations: identity, memory, trust levels, governance |
| openclaw-m2-memory-skill | Long-term memory for agents — hybrid semantic retrieval on Qdrant + BGE-M3 |
| ask-fable-skill | Delegate hyper-complex tasks to a Claude Code worker inside herdr |
Many other repos here are forks of open-source tools we run and patch.
- 🌐 machinemachine.ai
- 👤 Founder: Mariusz Kreft — AI strategy & engineering, grait.io
- ✉️ hi@grait.io