RKZ Agent Lab builds reliable multi-agent AI systems that operate real infrastructure — Android devices, homelab servers, and home networks — through controlled, verifiable processes. Powered by Claude.
Real systems, running on real hardware. Status labels reflect actual state — not marketing.
An always-on assistant with long-term memory, tool use, and multi-step task execution. Serves daily through Telegram, backed by a local model router with tiered fallbacks.
Our core research project: coordinated specialist agents with staged validation, privilege policies, and a single writer per target. Methodology first, automation second.
Agents performing verified operations on Android devices, Amlogic STBs, and network gear — every procedure tested, documented, and recoverable before it touches production.
How we work, on every project. Nothing ships on vibes.
Every operation starts with observation. We understand the system before we touch it.
Agents get the minimum capability needed. Root is a capability, never the default.
High-impact actions require a defined target, a backup, and a clear yes — per phase, not per project.
Every change ships with a rollback path and a documented last-known-good state.
Commodity hardware, open protocols, and frontier models — grouped by how we actually use them.
Building something with AI agents, or need automation for real systems? Let's talk engineering.