Updates, architecture notes, and release announcements.
Verified images, machine inventory, BMC power, and boot targets — how the provisioning dashboard makes a rack of servers tractable.
Declarative Kubernetes, Ceph SAN, object storage, and PostgreSQL on owned hardware — cloud-style workflows without hyperscaler markups.
How to compare Proxmox, Nutanix, OpenShift Virtualization, OpenStack, OpenNebula, and kcore when leaving VMware — and which operating model each one fits.
Build a Windows 10 Pro disk, create the VM with kctl, and open the desktop through an SSH port forward.
Inventory tools, disk movers, and Terraform generators are different jobs. Almost none of them write Terraform for a hypervisor that is not VMware.
Build the kcore ISO, turn on Hyper-V CPU enlightenments, create a Windows Server 2022 VM, and serve the default IIS page.
Contract end date, inventory, disk convert, and a per-socket quote — a VCSP exit plan that does not pretend cutover is a button.
Windows Server and desktop guests run. One GPU per VM. Live migration still needs Ceph. Not a Microsoft certification claim.
Shared Ceph for a VM move. Cold drain when you can stop. Multi-site control-plane replication is not vSphere HA.
A real Terraform provider for VMs, plus the 0.5.0 gaps: no Kubernetes, PostgreSQL, migrate, or GPU attribute.
A failed kcore host update boots the previous generation. That is the host, not a guest snapshot or a backup product.
VMs first. Kubernetes on the same metal when you declare it. One Cilium on the primary cluster. No surprise scale-in.
Per socket, not per core. Enterprise has an 8-socket minimum. Convert both quotes to a yearly figure before you argue features.
Proxmox fits a small UI-driven estate. Leave when the next step is an API, reviewable Terraform, and a tested convert.
Three jobs, not one button: inventory and a gap report, an exported-disk convert, then a cutover you can back out.
10% referral, reseller margin with deal registration, MSP per socket. White-label is the official bits, not a fork.
Two extra layers on top of normal tests: property-based testing with proptest, and bounded model checking with Kani. Where they apply in kcore (Nix escaping, path safety, controller database CRUD) and why we run both.
RFC1918 lab addresses, curl before and after, and a security group manifest that DNATs host TCP/8080 to an overlay guest running nginx.
A practical look at how kcore uses TLA+ for replication modeling, what classes of bugs model checking catches before production, where to learn TLA+, and examples of teams using it in real-world distributed systems.
A conservative, operator-focused walkthrough for early containerd support in kcore: create, list, inspect, start, stop, and delete containers directly on a node with clear caveats on current scope.
New quickstart with an embedded asciinema: install and approve your first node, add an operator SSH key, create a Debian 12 VM, and SSH in. Includes links to the new user docs at github.com/kcorehypervisor/kcore/docs/user.
Most distributed systems reach for Raft when they need replication. We chose CRDTs instead — here's how kcore uses a hybrid CRDT model to replicate controller state across datacenters without sacrificing availability during network partitions.
kcore started in Go. I rewrote it in Rust. No null errors, no GC pauses, a compiler that catches real bugs before they reach production — and it fits right into the NixOS ecosystem. Here's what pushed me to make the switch.
We've split the old monorepo and published the docs and website as separate public repos. The product docs are now at github.com/kcorehypervisor/kcore and the site deploys to kcorehypervisor.com via GitHub Pages. The core codebase stays private for now as we work towards the first public release.
Proxmox does a lot of things right, but its API is an afterthought, upgrades are scary, and the Perl codebase is showing its age. I built kcore to get declarative, API-first infrastructure on bare metal.