The manager’s Fabric Explorer — 256 components across 39 subsystems, mapped.
An autonomous infrastructure orchestrator for the Ring20 estate — it runs a private-cloud homelab end to end, directs a fleet of specialist agents, and gates every consequential action behind human approval.
The manager’s Fabric Explorer — 256 components across 39 subsystems, mapped.
Ring20 Manager is a maintainer-orchestrator agent — the lead of a small fleet — that runs a private-cloud homelab estate end to end. Its remit spans provisioning, monitoring, deployment coordination, and incident response, all under a task-governed agentic framework. The goal is an estate that largely looks after itself: routine operations happen autonomously, while anything consequential stops at a human-approval gate before it proceeds.
Rather than doing every job itself, Ring20 Manager directs a fleet of specialist agents — delegating provisioning, deployment, and observability work to the agent best suited to each, and coordinating the hand-offs between them. It operates inside the same governance model as the rest of the estate: nothing gets done without a task, and consequential actions are gated behind explicit human approval rather than trusted to agent discretion.
The agent is built on Claude (Opus) running under the Agentic Engineering Framework, and it talks to the rest of the fleet over a TermLink agent mesh. Underneath, it drives the estate’s infrastructure directly — Proxmox for virtualization, Cloudron for application hosting, and OneDev for source and CI — provisioning repositories, promoting deployments, and responding to incidents as they arise. It is a live, working part of the estate rather than a prototype.