We build the tools we wish existed — AI orchestration, real-time audio processing, and secure mesh networking — and run every one of them in production on our own hardware.
A private Claude Code launcher and Anthropic-compatible local proxy for running multiple model backends side by side. A unified web console layers on top: persistent terminal sessions, autonomy and permission controls, cost and budget tracking, an image prompt studio, and fleet-wide agent monitoring — all in one place.
A modular, real-time audio processing platform that turns commodity Linux hardware into a professional-grade guitar and audio rig. A C++ JUCE engine, Python FastAPI backend, and React dashboard combine to deliver neural amp modeling, convolution reverb, and a full effects chain at single-digit millisecond latency — with multi-node clustering over Ethernet for scaling a whole band on one shared digital backbone.
A secure, no-fixed-center workgroup mesh — and the egui-native, DRM-native desktop that runs on it. Up to eight machines form one private, encrypted workgroup with no central server: any node can author fleet policy, every node enforces it. On top sits a thin client that brokers full VM desktops — Windows, Ubuntu and more — rendered natively over the mesh, all on a Nebula overlay with etcd coordination and Syncthing replication.
Also from the studio: MDE — Mackes Desktop Environment · featured on the résumé →
Every product here started as an internal need, not a pitch deck. We design, build, and operate each one ourselves — which means the code you find in these repos is the same code running our own infrastructure today.
ephemeral.team is the independent studio of Matthew Mackes — AI risk & governance architect in Buffalo, NY, with 25+ years securing and shipping production systems. Read the résumé →