Burgward.org
Work in progress. A proof-of-concept for fully reproducible base images exists; the catalog and signing chain are still in design. First images to launch: postgres · redis · node.js · python.
Proposal · evaluation phase · June 2026

Pull every image.
Verify the chain.

Hardened, Wolfi-derived base container images — designed to be always free and always the latest version, with European sovereign roots and a full signature chain. Each image will ship a reproducible build, a CycloneDX SBOM, in-toto provenance and a cosign signature from an EU-resident key — so you re-derive and match the digest rather than trust us.

The problem

why this exists
Everything runs on a base image — and the base layer is the supply chain's weakest, highest-blast-radius point.

EU institutions and businesses wanting to adopt sovereign open-source software hit a wall: the container images they're handed can't pass security review — standard bases (e.g. Debian) carry hundreds of open CVEs, with no reproducibility, provenance or signing on the base layer, and a single flaw in a base component cascades into every image built on it.

And there is no EU-rooted hardened alternative to turn to. That unmet need — a base layer European institutions and businesses can actually trust and verify — is what Burgward offers.

Why Burgward

vs. existing hardened-image options

Hardened base images already exist — but every credible one is US-rooted, so none can be the sovereign root of trust for European institutions and businesses under the CLOUD Act. They are the upstream we build on and contribute back to, not a substitute.

DistrolessGoogle — US. Generic infra; trust rests on Google's signature; build bootstrap not re-rooted.
ChainguardUS company (free + commercial). Trust = the vendor's signature, not independent re-derivation; no EU signing chain.
BitnamiBroadcom / VMware — US. Hardened images, but the same jurisdiction and signature-trust limits.
BurgwardEU-rooted. Built from source with our own keys; trust by independent re-derivation — rebuild and byte-match the digest, no vendor to believe; bootstrap re-rooted. Apache-2.0, contributed upstream to Wolfi.

Status

done vs planned · 2026-06-01

✓ Done so far

  • Reproducible-build proof-of-concept: a Wolfi-derived hardened base rebuilt byte-for-byte from source, with our own signing key — Apache-2.0 tools only, no vendor binary in the trust root.
  • Design & architecture of the catalog, signature chain and verifier.
  • Codeberg organisation (EU-hosted source).

Free at the root.
Forever.

The base layer is what every container is built on top of. It should be a commons — open, verifiable, and free — not a layer you rent from a single foreign vendor. The intent: the images and their machine-readable proofs will live here, openly, under a free license, maintained continuously.

The scope is deliberately narrow — base images and the proofs that travel with them: recipes, SBOMs, in-toto provenance and cosign signatures. Reproducible, EU-rooted, free to pull and modify.

scopeImages · recipes · SBOM · provenance · cosign signatures · continuous rebuilds planned

Base-image catalog

First four images · in evaluation
postgres
on burgward-minimal · PostgreSQL License
18.4In evaluation
$ docker pull images.burgward.org/postgres:18.4planned
SBOMPROVENANCESIGNEDREPRODUCIBLE
redis
on burgward-minimal · AGPLv3
8.8.0In evaluation
$ docker pull images.burgward.org/redis:8.8.0planned
SBOMPROVENANCESIGNEDREPRODUCIBLE
node.js
base image · on burgward-minimal · MIT
24.16 LTSIn evaluation
$ docker pull images.burgward.org/nodejs:24.16.0-ltsplanned
SBOMPROVENANCESIGNEDREPRODUCIBLE
python
base image · on burgward-minimal · PSF
3.14.5In evaluation
$ docker pull images.burgward.org/python:3.14.5planned
SBOMPROVENANCESIGNEDREPRODUCIBLE

Planned base: burgward-minimal — a glibc, distroless base to be rebuilt from Wolfi source with our own EU keys. Component images for specific sovereign stacks would follow. Versions shown are evaluation targets and may change.

Trust by re-derivation, not by signature alone. Planned model

A signature only says we built it. The intended model: because every image will be built reproducibly from a declarative source recipe, anyone — you, an auditor, another EU operator — can rebuild it and confirm the digest matches, bit for bit. The trust root is the source and the build, not a vendor you have to believe. A planned sovereign-verify CLI would check the whole chain offline, against a bundled trust anchor — for air-gapped networks too.

01 · source
Recipe
Declarative melange/apko, public on Codeberg.
02 · build
Reproducible
Pinned, deterministic — byte-for-byte re-derivable.
03 · sbom
CycloneDX
Full component inventory, per image.
04 · prov
in-toto
SLSA provenance — what built it, from what.
05 · sign
cosign
EU-resident HSM key. No US trust root.
CycloneDX SBOM in-toto / SLSA provenance cosign signature · EU HSM Reproducible build all planned

Designed to stay latest.
Rebuilt continuously. Planned

Compliance decays with every CVE, so the commons is designed not to freeze. The plan: when a vulnerability lands in a base component — the Dirty Pipe / xz-backdoor class, whose blast radius reaches every downstream image — the affected images are re-derived, re-attested and republished. Always-latest as the default.

advisory → re-derive
→ rebuild from source
→ re-attest (SBOM·prov·sig)
→ republish
→ you re-verify

European sovereign roots

the design · no US vendor in the trust root
From source

Rebuilt, not mirrored

To be built from Wolfi source recipes with our own generated keys — not re-signed vendor binaries. The bootstrap is re-rooted so trust starts from source you can read. (Proven in the proof-of-concept; production catalog planned.)

EU-resident

Signed in Europe

Planned: cosign signatures from an EU-resident HSM key; source on Codeberg (EU). No CLOUD-Act jurisdiction over the root of trust — structural, not a setting.

Given back

Upstream & open

Recipes to be Apache-2.0 and contributed upstream to Wolfi. The commons grows the shared stack rather than forking it — the patina, not another wall.