How to use Codex to build a homelab control plane
reactordev used Codex to build a Bun and React management plane for Linux GPU machines, with privileged Docker agents, WebSocket reporting and project-based delegation.
news.ycombinator.com
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Build a management plane for [Linux GPU machines] using [UI stack]. Start from [protobuf protocol specification], add authenticated access and SSH-based installation of Docker-management agents, and implement WebSocket health reporting plus project repositories, artifacts and prompt histories.
Reach people exploring AI agents
See placements →What happened
The author reports building the initial system in roughly a month with Codex, then moving coding work to a custom agent and finding project-based delegation effective.
Reported by reactordev on Hacker NewsHow the agent did it
- Defined the communication protocol upfront using protobuf.
- Built a Bun management plane with a React UI and authentication.
- Added SSH installation of privileged Docker-management agents.
Where a person stepped in
The author defined the protocol and system structure; the post does not describe a formal code-review process.
A person approved each step before the agent continued.
reactordev used Codex for most of the initial work on a management system for Linux machines with GPUs. They already had a Raspberry Pi serving as a homelab Pi-hole and DNS host, and chose it to host the management UI. The interface was a small Bun management plane with a React application, user authentication and OpenID Connect.
Before implementation, the author defined the protocol using protobuf, even though communication used WebSockets. They say this helped preserve the vision and keep Codex on track through completion. They then added direct SSH access to install a privileged agent capable of managing Docker. Installed agents reported health, status and resources every 15 seconds, with streams for logs or container access.
The system modeled workloads through organizations and projects. Projects held Git repositories, image registries, artifacts and prompt histories. Agents assigned to the same project could share a stream, while project URLs exposed interfaces for interacting with project scrum masters. The initial Codex build took roughly a month, before the author's custom agent became capable of coding against a configured endpoint.
Who it's for: Experienced platform engineers exploring homelab resource pools and project-based agent delegation.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · reactordev
Frequently asked questions
Why define protobuf before coding?
The author says defining the protocol upfront helped clarify the vision and keep Codex on track through completion.
How do installed agents report status?
They communicate with the management plane over WebSockets and send keep-alive, health, status and resource packets every 15 seconds.
More details
Remaining steps
- Modeled organizations and projects with repositories and prompt histories.
About this account
Codex performed most of the initial implementation, guided by a protocol the author defined upfront.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=48454527. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: reactordev on Hacker News
Source published 9 Jun 2026
Source checked 14 Oct 2026
Tools and topics
Codex · Bun · React · protobuf · Docker · SSH · WebSockets
Homelab · Agent orchestration · Docker
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.