About

Knowing you have the part is half of quoting a repair. bench-parts is the inventory that answers it — stock levels, what got consumed on which job, and a to-order list you check before a supplier run instead of discovering the gap mid-repair.

It has two faces, and neither depends on the other:

  • A standalone app — a tablet-friendly web UI for the bench. Add a part, adjust a count when you restock or use something, check what’s running low. Works fine with nothing else running.
  • A parts REST API — deliberately consumer-agnostic. Any authorized client on the LAN can read stock and record consumption. benchhud is the first consumer through an inventory adapter, but a dashboard or a shell script could hit the same endpoints with no server-side change.

The design constraint that shapes it: no customer data lives here. Parts are stream-safe. That’s what lets benchhud put inventory on the HUD — “you’re on job 26-0001, low on IRF7910” — while it’s compositing to a live stream. Consumption links to a job by reference string only, and this service never calls intake to resolve it. The boundary is the point.

It’s grown a second register, too: test gear. An asset catalogue for the instruments themselves — check a scope or a probe out, check it back in, and see where everything is — with the same JSON API. And printable QR drawer labels, with a CLI that sends them straight to a Niimbot D110 label printer.

Python and FastAPI, SQLite. It installs as a .deb service on 127.0.0.1:8402, a tray app, or a container; mine runs on a Raspberry Pi cluster over Flux. Multi-arch images, with CI asserting the published manifest actually advertises linux/arm64 — an amd64-only build is invisible until the pod refuses to schedule.

Open source under AGPL-3.0-or-later — use it, change it, run it. If you run a modified build as a service other people use, your changes go out under the same license.