🏗️ Infrastructure — the 5 layers beneath Quilt

The 5 new infrastructure repos that make Quilt solid, predictable, and resilient. Designed with z.ai and Kimi.

Home Workspace Playground Fleet Chaos test Quilt+Nomad Infrastructure Pincher Vibe

Every solid platform has layers beneath the surface. The 5 infrastructure repos below are the foundation of Quilt: minimal container images, two embedded orchestrators, an immutable appliance, and a chaos testing framework. They were built from a conversation about making orchestrators structurally solid, predictable, and resilient. Each one has a splash image, a clear mission, and a working demo.

The shared philosophy

All 5 repos share three principles. They're not just technical choices — they're the design values that make Quilt feel like one platform instead of 25 loose tools.

🎯 Zero-config defaults

Every repo works out of the box with no setup beyond `npm install`. The defaults are the right choices for 90% of users. Power users can override anything, but they never have to.

🛡️ Solid by default

Each repo is designed to be hard to break. Minimal CVE bases. Transactional updates. Chaos-tested recovery. Immutable appliances. Resilience isn't an afterthought.

🔌 Composable everywhere

Each repo can be used standalone or composed with the others. quilt-base runs on quilt-core-os. quilt-k3s tests quilt-agent. The whole is more than the sum of parts.

The 5 infrastructure repos

quilt-base

The foundation layer. Every Quilt agent runs on a quilt-base image.

Minimal, hardened, auto-rebuilding, zero CVE. Four container variants (Alpine, Ubuntu, Distroless, Scratch) — 4MB to 28MB. Daily CVE scans with Trivy and Grype, SBOM generation with Syft, image signing with Cosign. Multi-arch (amd64 + arm64). Non-root user. The base you don't have to think about.

GitHub README Dockerfiles Security scripts

quilt-swarm

Quilt as a control plane for Docker Swarm. Edit a cell, the cluster reconfigures.

Encrypted overlay networking, service mesh, secret rotation without downtime. Map a Quilt cell to a Swarm service: value→count, formula→service spec, api→health check, listener→handler. 28 tests passing. Use this when you only need containers and want minimal infrastructure complexity.

GitHub README Source Tests (28) Live demo

quilt-core-os

An immutable Ubuntu Core appliance. The OS you cannot break.

Read-only filesystem. Cryptographically signed system. Transactional OTA updates with automatic rollback. Survive a power loss mid-update. The kernel, the OS, the agent — each is an independent layer that can be rolled back without affecting the others. Pi 4, Pi 5, Jetson, x86. The edge device you can deploy and forget.

GitHub README Snap Gadgets Docs

quilt-k3s

Chaos engineering for Quilt. 5 failure scenarios. Self-healing verification in under 30s.

Spins up a 3-node K3s cluster in CI. Injects failures. Verifies recovery. The 5 scenarios (node failure, network partition, disk full, API down, etcd down) and recovery thresholds are designed by Kimi moonshot-v1-8k. 34 tests passing. Catches resilience regressions before they reach production. The immune system of the Quilt platform.

GitHub README Scenarios Tests (34) Live demo

quilt-nomad

Quilt as a control plane for HashiCorp Nomad. Cell → job, formula → task, api → http check.

Same idea as quilt-swarm, but for Nomad's multi-task scheduler. Containers, raw binaries, Java JARs, systemd services — all from one Quilt sheet. 6 Nomad API endpoints wrapped. Cell-to-Nomad mapping designed with z.ai GLM-4.5. 14 tests passing. Use this when you have a mixed workload on a small cluster.

GitHub README Source Tests (14) Live demo

How they fit together

The 5 repos aren't 5 separate projects — they're 5 layers of one platform. quilt-base is the image. quilt-core-os is the appliance that runs the image. quilt-swarm and quilt-nomad are the orchestrators that schedule it. quilt-k3s is the chaos test framework that validates the whole stack. They cross-reference each other. The README of each one links to the others. They use the same engineering bar (Apache 2.0, GitHub Actions CI, CODEOWNERS, Dependabot, ESLint). They share the same philosophy (zero-config defaults, solid by default, composable everywhere).

The 5 layers form a self-reinforcing loop. The base image makes deployments predictable. The orchestrators make the deployments declarative. The immutable OS makes the deployments durable. The chaos tests make the deployments verified. Every layer makes the other layers possible.