Chainguard surpasses one billion container build manifests

Chainguard doubles build-manifest output in six months

Chainguard says its Factory produced more than 1 billion container build manifests, up from 500 million over the past six months. The company also reported a catalog of more than 3,000 unique container images and 675,000 image versions.

In Chainguard’s definition, a build manifest is created whenever the factory generates a new verifiable artifact. That can include a fresh image such as go:1.26.5, an nginx rebuild prompted by a libc patch, an architecture-specific variant, or a regenerated software bill of materials after a dependency change.

The metric therefore reflects repeated rebuilding across supported projects as upstream packages change, dependencies are patched and base images are hardened. A project such as Python can span multiple supported versions and architectures, each requiring further builds over time.

Factory 2.0 shifts from event handling to reconciliation

The catalog is built on Chainguard OS, a rolling-release Linux operating system intended for cloud-native workloads. Chainguard says new artifacts ship throughout the day rather than through infrequent distribution releases. Factory artifacts are built from source and include SLSA Level 3 provenance, Sigstore signatures and full SBOMs.

The original Factory automated package definition, dependency resolution, builds, signing and delivery. As the catalog grew, Chainguard says the event-driven design created brittle queues, duplicate failures, work-item conflicts and cases that needed manual intervention.

Factory 2.0 introduces DriftlessAF, an open-source agentic framework that continuously compares a defined target state with the actual state. It assigns reconciler bots work from a shared queue using information from code repositories, security feeds and other sources. A reported CVE, upstream package release or new policy criterion can create a gap for the system to close.

Chainguard describes the model as redundant by design: a failed task can be retried or dropped because the system is working toward an end state rather than relying on every individual action succeeding. AI is used for unstructured decisions, including evaluating newly added components in minor releases and backporting CVE remediations to older package and language releases, while structured and verifiable tools constrain the process.

Rebuild velocity becomes a supply-chain control

Chainguard links faster rebuilding to a threat environment in which AI can accelerate vulnerability discovery, exploit development and exploit chaining. The company argues that reducing the interval between an upstream change and a rebuilt, signed and verified image narrows the window in which an affected artifact remains available.

For businesses consuming container images, the practical implication is to assess not only an image’s initial security posture but also the supplier’s ability to rebuild, document and distribute updated artifacts as dependencies and vulnerability information change.