Spergs, LLC

← All products

Backplane

In progress

Every machine, accounted for.

About Backplane

Backplane exists because a fleet of a dozen machines had a dozen sources of truth: a README per box, an inventory per project, a monitoring stack that knew some of them, and the memory of whoever last logged in. Now each machine has a declared record in its own git repo, and Backplane reads it, lints it, and compares it field by field against what the machine reports about itself. The disagreements are the product — a machine that says it is on one kernel and runs another, an address that moved, a declared GPU driver that is not the loaded one.

An observe-only agent on every machine reports its own facts and never runs a command. Health is judged against what each machine is supposed to do, so a laptop that is asleep is asleep rather than an outage, and an always-on box that goes quiet pages within minutes. Appliances that take no agent are read through their own APIs instead: the NAS reports its pools, and a degraded one is a finding with the failing disk named.

Changes go through one controller. A machine says in its own repo which roles it takes; the controller generates the play, runs a check, and applies only what a successful check covered — then writes the result into that machine's changelog in git. Runs drain around each other, a machine can be taken out of rotation, and secrets reach a run as files the deploy wrote, never as values the play can print. Six machines in the fleet converge from git this way today.

It is a set of small C++ services on Boost.Beast built from vendored source, with Postgres underneath, deployed by Ansible and systemd, no Docker. It comes with a dashboard, a command-line client, an MCP server an agent session can drive, and an Android app that receives health pushes.

What it is not, named here rather than implied: there is no hosted Backplane — it runs on a machine you own, on a network you trust — and no licence has been chosen, so it is not offered for download today. The Android app is signed and delivered by the build service to one handset; it is in no store.

At a glance

  • Declared machines read from git, linted, and diffed against what they report
  • An observe-only agent on every machine; it never runs a command
  • Health judged against each machine's availability — asleep is not an outage
  • Appliances read through their own APIs: a degraded NAS pool names its disk
  • One controller applies changes: check, then apply what the check covered
  • Every apply written to that machine's changelog in git
  • Dashboard, CLI, an MCP server, and an Android app with health pushes
  • Self-hosted C++ on Ansible + systemd, no Docker

More from Spergs

Other independent software we ship, or are working on.