Apple Intelligence tools shared

A pair of small posts this week made a blunt, practical point: Apple’s on‑device AI is no longer a promise for system features alone — people are starting to use it. (x.com 1) (x.com 2) One post came from a designer who shared a short motion‑design demo built with Paper’s new “shaders” — tiny, exportable canvas effects that can be composed in a design tool and then dropped into a website or app. The demo shows colorful, procedural textures and animations that were exported as runnable code, not just a video. (github.com) (shaders.paper.design) The other post pointed to a GitHub project that turns the on‑device model Apple ships into a command‑line tool and a simple HTTP server. The repo advertises a lightweight CLI named “apfel” that wraps Apple’s FoundationModels APIs so developers can call the same on‑device language model from shells, scripts, or their own servers. The project emphasizes that every inference runs locally on the Mac — no cloud keys, no token bills. (github.com) Those two artifacts are small by themselves. A shader component and a community wrapper are not company releases. What they show, though, is different: builders are already experimenting with Apple Intelligence in practical ways. The shader demo maps to designers’ workflows — create in a visual tool, export code that actually runs — so it shortens the path from idea to interactive prototype. (github.com) The CLI project shows the same pragmatic impulse on the developer side. Apple exposes the on‑device model through the FoundationModels framework; apfel makes that framework usable from places Apple does not officially expose it — the terminal and a local HTTP endpoint — so scripts and server processes can include on‑device LLM calls. The result looks like a local, privacy‑first alternative to a cloud API: your Mac does the inference. (developer.apple.com) (github.com) How this works in plain terms: Apple packages a language model inside the operating system and provides a framework named FoundationModels that lets apps start a session with that model and ask it to generate text or call simple “tools.” Developers can call those APIs from Swift and, thanks to community projects, from the shell or local servers as well. Because the computation runs on Apple silicon in the device, user data does not have to be sent to a remote service. (developer.apple.com) Why these posts matter now is practical and immediate. Design teams can show motion and interactivity driven by real model outputs rather than mockups. Engineers and sysadmins can script local LLM calls into existing workflows without billing surprises or network dependencies. Together those two threads shrink the friction between an on‑device model and actual products or prototypes. (github.com 1) (github.com 2) If you want a concrete next step: the apfel repository includes usage notes and claims compatibility with macOS versions that support FoundationModels, so a developer with an Apple Silicon Mac and macOS that supports Apple Intelligence can install the tool and run a local server to experiment with LLM prompts and file attachments right away. (github.com)