Machin in Production — Real Products Built with MFL

When I first wrote about machin — the machine-first language that compiles through C to a single native binary — the most common question was: "OK, but what can you actually build with it?"

Fair question. A language that optimizes for AI generation over human readability sounds good in theory, but theory does not ship products. Here is what we have actually built with machin, running in production today.

machin-auth — Self-Hosted Auth Broker

A single binary (~80 kB) that handles Google OAuth sign-in, issues EdDSA-signed identity tokens, and exposes a JWKS endpoint any application can verify against. No database, no session store, no external dependencies. Deploy it behind any reverse proxy, configure a client ID, and every application on your domain gets instant, verifiable authentication.

This replaced a Node.js auth service that weighed 200+ MB with dependencies. Same functionality, three orders of magnitude smaller, zero runtime.

machin-hook — Webhook Relay + Live Inspector

A self-hosted replacement for webhook.site and smee.io. Receives webhooks, inspects them in a live terminal UI, and relays them to your local dev server. One ~80 kB binary, no Docker, no Node. Runs on any Linux machine, edge deployments included.

Every developer on our team uses it daily during integration work. It replaced a Docker Compose setup with a single sc machin-hook command.

machin-playground — Browser REPL

A WebAssembly-compiled version of machin that runs in the browser. Write MFL code, hit Run, it compiles to WASM and executes client-side. Zero server roundtrip. This is how potential users try the language without installing anything.

The entire playground is a single machin binary that serves both the API and the static frontend.

machin-terminal — Minimal Terminal Emulator

A correct, fast, single-window terminal emulator in pure MFL. One static binary, no GTK, no VTE, no Node. Pairs with tmux for multiplexing. Proof that machin can handle raw system programming — PTY management, terminal escape sequences, signal handling — all in a language designed for agents.

The Pattern

Every product above follows the same pattern: one static binary, zero runtime dependencies, minimal attack surface. They deploy via copy-paste, not Docker pull. They start in milliseconds, not seconds. They work on a $5 VPS just as well as on a Raspberry Pi.

This is what agent-first infrastructure looks like: tools so small and self-contained that an AI agent can deploy, configure, and manage them without human intervention. No package managers, no dependency hell, no "works on my machine."

machin on GitHub — v0.53, MIT, one binary to start building.

Next in this series: a2a — when AI agents talk to each other.