bty-lab 0.28.0

Flash images onto target disks, offline or networked with and without PXE

bty - flash images onto target disks, offline or networked with and without PXE

Pronounced "battie" (rhymes with "batty") - the blue bat up top is the mascot, so when in doubt say it like the critter.

Flash a single bare-metal box ad-hoc with a USB stick, or reflash a whole fleet remotely from a single controller -- bty works with or without PXE and scales from one machine to a rack without changing how you operate. The image is the source of truth: rebuild the image, reflash the target. No imperative configuration management, no idempotency mind games. Works equally well in homelabs, CI fleets, lab benches, data-centre racks, and anywhere else bytes need to land on a disk.

bty is a flasher, not an image builder:

• Image creation is somebody else's project. First-boot bring-up (users, network, packages, hostnames) gets baked into the image upstream with cloud-init / kickstart / preseed / your favourite image builder. Use the companion image-builder (safl/nosi -- builds Debian / Ubuntu / Fedora sysdev images and publishes them to GHCR as ORAS artifacts that bty flashes via oras://), or your own. bty just writes the bytes.
• No post-boot configuration management either. Anything that needs to be true on the running target (users, hostnames, config files, packages) belongs in the image builder, not in bty. The server does not hold creds for any target it has provisioned -- that blast radius is intentionally absent.

# Local: USB stick into target, two arrows + Enter, done.
bty

# Remote: bind a MAC to an image, the next PXE boot reflashes itself.
# (See the bty-web HTTP API reference in the docs for the full surface.)

# Per-job CI: every job a clean OS, no drift, no snowflakes.

Three delivery shapes, one runtime

Shape	What it is	When it fits
USB live stick	bty boots from a flash drive, runs bty, flashes the box it's plugged into. Fresh sticks ship with four starter .bri pointers (Debian / Ubuntu / Fedora sysdev images via oras://ghcr.io/safl/nosi/..., plus bty-server) so the catalog is non-empty out of the box.	Single-machine local imaging
USB + portable catalog	Same stick, plus bty --catalog <SOURCE> pointed at a TOML catalog hosted anywhere (a local file, an HTTP URL, an oras:// reference, or a bty-web instance's /catalog.toml).	A handful of boxes, shared image library
PXE-boot appliance	bty-web on a Pi or x86 box runs DHCP/TFTP/HTTP; targets PXE-chain into a netboot live env that runs bty --server X --mac Y on tty1, which fetches a per-MAC plan and either auto-flashes or drops the operator into the wizard	CI fleets, racks, anything you don't want to walk to

All three share the same Python codebase, the same image catalog, the same SHA-keyed machine bindings.

The PXE-boot appliance also separates rootfs from image cache: drop a 2nd disk in, run sudo bty-image-store-init /dev/sdX once, and the image library survives appliance reflashes. The new appliance auto- mounts the labelled disk at /var/lib/bty/images; no operator action required.

ORAS-published images and portable catalogs

bty consumes images and catalogs as OCI artifacts published with ORAS (OCI Registry As Storage -- the spec for non-container artifacts in a container registry). The end-to-end story:

• Images live in a registry. safl/nosi publishes Debian / Ubuntu / Fedora disk images to ghcr.io/safl/nosi/<variant>:latest. bty resolves an oras://ghcr.io/safl/nosi/... source from a catalog entry, picks the disk-image layer, and streams the blob straight to the target via the same curl | dd pipeline as any HTTP URL. Anonymous-pull only -- no PAT, no docker login.
• Catalogs are portable TOML files. A catalog is a small TOML manifest listing named images with src URLs (any combination of http(s)://, oras://, or file://). bty --catalog <SOURCE> accepts a local path, an HTTP URL, or an oras:// reference. Operators can publish a catalog on GitHub Releases, an S3 bucket, a private registry, or alongside images in GHCR -- whatever they already have. bty-web instances serve the same shape at GET /catalog.toml, so a running server is "just another catalog source".
• .bri descriptors are the per-stick analogue. A USB stick's BTY_IMAGES partition can carry .bri files (one-image-per-file TOML pointers, including oras:// URLs). The TUI merges them with whatever --catalog source the operator passed.

Why this shape: images and catalog metadata are content-addressed artifacts, not container images. The OCI ecosystem already solves "distribute signed, versioned, content-addressed blobs"; bty just piggybacks on that without dragging in the docker / podman runtime.

Why bty

• Reflash on every CI job. Per-job cadence: each job lands on a freshly-imaged target, runs, gets reflashed for the next job. No state leaks. No snowflakes. No "works on my machine" because the machine is bit-identical to the manifest every single boot.

• Pre-built images, not recipes. You build the image once (in your build system of choice), bty writes the bytes. Any first-boot bring-up (users, networking, hostnames) is baked into the image by the image builder upstream via cloud-init / NoCloud user-data. bty itself doesn't run a provisioning step -- no agent, no daemon, no convergence loops.

  Note that interactive picks (operator chooses an image at tty1) are not reported back to the server: bty-server tracks "what image is this MAC supposed to have" only when a flash policy (boot_policy=bty-flash-always / bty-flash-once) binds it. Interactive runs are operator-driven and stay local. See docs/src/concepts.md for the asymmetry.

• OS-agnostic by design. Linux, FreeBSD, Windows - if it boots from a disk image, bty can flash it. macOS targets are out (Apple Silicon's boot story isn't friendly to imaging).

• Trust model is explicit. PXE / live-env routes are open (clients have no token); operator routes (/machines, /catalog/*, /boot/releases) require a session cookie. bty-web is for trusted networks (homelab, CI segment), not the open internet.

Try it without flashing anything

A multi-arch container is published on every release:

docker run -d --name bty-web -p 8080:8080 -v bty-data:/var/lib/bty \
  ghcr.io/safl/bty-web:latest
# -> http://localhost:8080/ui   (login: bty / bty)

HTTP-only - no TFTP daemon bundled in the container. The container's lane is UEFI HTTP Boot (operator's DHCP serves option 67 = http://<bty>:8080/ipxe.efi) or pairing with a boots-from USB stick (operator boots the stick, embedded iPXE chains to bty's HTTP endpoint). For fleets that need TFTP (legacy BIOS + UEFI firmware that only does TFTP option 67), use the bty-server appliance -- it bundles dnsmasq for TFTP serving alongside bty-web. See docs/src/walkthrough-server-docker.md for bind-mount permissions, env vars, and password rotation.

Install

bty is one Python package - bty-lab on PyPI - with two console scripts:

pipx install "bty-lab[tui]"     # `bty` (Rich-based wizard, the
                                #  operator-facing tool)
pipx install "bty-lab[web]"     # adds `bty-web` (FastAPI + Pydantic,
                                #  the controller appliance)
pipx install "bty-lab[all]"     # everything

bty shells out to dd, qemu-img, zstd, lsblk, curl (used by URL / oras:// fetch), and friends - your distro provides those. The dispatch surface is intentionally narrow: bare bty launches the local-image wizard, bty --catalog URL pre-loads a catalog, bty --server X --mac Y fetches a per-MAC plan from the server and dispatches (auto-flash / interactive / no-op).

For an appliance you can boot directly (USB stick, server image, PXE-chain live env), grab the bake from GitHub Releases. The appliance builder lives under bty-media/.

Status

Pre-1.0 but actively shipping. Every tag publishes wheels (PyPI), appliance images, and the bty-web container. The end-to-end PXE flow (server + netboot live env + target flash + completion signal) runs in CI on every push. CLI flags and wire formats may still shift between minor versions until 1.0 - watch the schema_version field on --json output and the Machine wire type. The PLAN.md tracks the roadmap milestone by milestone.

Development

pipx install uv
uv sync --all-extras --group dev
uv run pytest                    # full suite
uv run ruff check                # lint
uv run mypy src                  # types

The docs tooling installs separately:

pipx install ./docs/tooling
cd docs
bty-docs-serve                   # live-rebuild dev server on :8000
bty-docs-build-html              # one-shot HTML build
bty-docs-build-pdf               # one-shot PDF (requires LaTeX)

More

• PLAN.md - roadmap and design intent.
• docs/ - full documentation (Sphinx + MyST), also at safl.dk/bty.