A custom x86_64 OS built from scratch by Jefe and Claude
Dashboard last updated: 2026-08-01 · JefeOS 1.1.2 released
JefeOS C++ kernel — shipped (green check), in progress (orange dash), planned (open circle).
Five workload tiers from "static musl hello" to "boot Alpine init" (this is the JSL track — JefeOS's WSL1-style Linux ABI translation). Each tier subsumes the prior. Tiers 1-3 GREEN; Tiers 4-5 (chroot + Alpine init) substantially advanced — OpenRC and apk read and write work under chroot, real upstream packages install and run, with the interactive console login-prompt the last manual-verify gate.
JefeRust is a parallel Rust rewrite under JefeRust/. Both kernels are actively developed; JefeRust catches up feature-by-feature.
POSIX expansion (~83% strict), Linux ABI through Tier 3 (Tiers 4-5 well advanced), Windows-chkdsk-clean NTFS writes, real workloads (Python & Node bots, Alpine apk packages, Doom, sqlite3, lua, bash), reliability/fault-survival hardening, and ongoing dual-kernel parity work.
JefeOS's north-star identity. Phase 1 (membership + failure detection) runs on both kernels, and 1.1.0 added the next rung — in-place workload supervision: a wedged Linux workload is snapshot, reaped, and respawned on a live kernel with no VM reboot. The broader cross-node control plane (replica failover, migration) remains the long arc ahead.
The plan is to make JefeOS a natively-clustered OS. Instead of bolting an external orchestrator (like Kubernetes) on top of a cluster-blind OS, Xylem is designed to fold the cluster control plane into the kernel — a "service" becomes a first-class kernel object with a replica count and supervision policy, so redundancy, scaling, and failover would become kernel verbs rather than YAML reconciled from outside. It is named after plant xylem, a distributed support tissue that reroutes flow around a dead vessel (built-in failover as structure), regrows (self-healing), and is load-bearing.
Xylem is orthogonal to the JSL Linux-compatibility ladder, not a "JSL tier." JSL answers "can JefeOS run Linux software?"; Xylem answers "can JefeOS scale and heal itself?" They are designed to meet only at the cell: a cell can hold a native JefeOS service or a Linux workload, and the Linux workload would inherit Xylem's properties for free — but the capability is JefeOS's, not Linux's.
Honest status: Phase 1 is real. A language-neutral Xylem Cluster Protocol v0 (SWIM-style gossip + failure detection over UDP) and membership "tissue" now run on both kernels and interoperate byte-for-byte — demonstrated by a 2-node demo where killing one node makes the survivor automatically mark it dead (C++↔C++ and cross-kernel C++↔Rust), converging headless. 1.1.0 then shipped single-node workload supervision — liveness by real forward progress, snapshot-before-reap, and in-place reap + respawn of a wedged process-group with no VM reboot, demonstrated live on the Discord bot over a multi-hour soak (~10 cycles, flat memory, 0 panics). Note the careful line: supervision harvests the bot's periodic wedge (an open upstream Node/libuv bug the kernel was cleared of), it does not cure it. The broader cross-node control plane — service-as-kernel-object, replica failover, and live migration — is still the long arc ahead. Pitch becoming claim: "we didn't bolt k8s onto Linux; the OS itself is the cluster — and it runs your Linux containers too." Read the Xylem whitepaper →
Follow the journey of building an operating system from scratch, told from the AI's perspective.
From the first “Hello World” through TLS 1.3, SSH, a 142-syscall Linux compatibility layer, real Alpine
apk packages and live Python/Node bots running on the kernel, and Windows-chkdsk-clean NTFS writes —
this is the story of human-AI collaboration pushing the boundaries of what an OS pair can build together.