01
Linux as a host — and macOS containers on Linux
Designed · seams in place
Today dd hosts on macOS. Next: the same engine running on Linux — Linux and macOS containers, on a Linux box — so the host stops mattering. The runtime is already split along host/guest seams (HAL, per-ISA emitters, OS personalities); the work is wiring the Linux host primitives and the Mach-O/Darwin guest path on Linux.
Why it matters: run your Mac's containers on a Linux server, and Linux's on your Mac — one runtime, any host.
02
Full rendering on macOS — Metal access + a universal render bridge
Designed
Let containerized apps drive the GPU through Metal, and ship a rendering bridge so any app — Linux GUI, game, creative tool — paints to a native macOS window. A guest's graphics calls map onto Metal/Core Animation instead of a remote framebuffer.
Why it matters: GUI and GPU workloads, not just headless services — at native frame rates.
03
Secure, declarative dev environments — perfect DX
Planned
Describe what you need in a config file and get a reproducible, sandboxed environment in seconds — the right toolchain, services, and isolation, on any machine. Untrusted code runs behind a real trust boundary (the sentry split), so "just try this repo" is safe by default.
Why it matters: it should be trivial to work on anything, anywhere — no "works on my machine," no VM tax.
04
A cross-platform terminal emulator
Planned
An iTerm2-class terminal built on dd — but able to drop you into a shell for any OS and architecture the runtime emulates, with the same UX everywhere. Interactive TTYs already work end-to-end (real pty, signals, resize); the terminal makes that the front door.
Why it matters: one terminal, any OS/arch shell — local, containerized, or cross-platform.
05
First-class agents & LLMs — optimized CUDA/Metal bindings
Researching
Run LLMs and agent workloads of any kind, with optimized CUDA and Metal bindings so model code reaches the GPU at full speed — whether it was written for one accelerator or the other. The container brings the model; dd bridges it to whatever silicon is present.
Why it matters: portable AI — the same model and agent stack runs on a Mac's Metal or a server's CUDA.
06
Native performance everywhere — the throughline
In progress · continuous
The mission under all five pillars: an app should run as if it were compiled for the machine it's on, no matter the source OS or CPU. Straight-line compute already runs at — and sometimes above — native; ongoing work closes the remaining gaps (indirect-branch dispatch, syscall and I/O paths, GPU bridging) so cross-arch isn't a tax.
Why it matters: portability with no performance penalty is the whole point.
dd