Developer Arthur-Ficial has released apfel, a free open-source tool that unlocks Apple's on-device LLM for Mac users running macOS 26+. The project, which has already earned over 200 GitHub stars, provides three interfaces: a CLI tool, an OpenAI-compatible server, and an interactive chat, all running entirely on Apple Silicon hardware without network calls or API keys.

Apple ships its language model as part of the Apple Intelligence framework, but restricts access to Siri and system features. apfel bridges this gap by wrapping the FoundationModels framework, exposing the 4,096-token model through familiar interfaces. Users can install it via Homebrew with a single command, then pipe text through the model using standard UNIX tools or point any OpenAI-compatible SDK at localhost:11434 for seamless integration.

The tool offers practical features like five context-trimming strategies for the limited token window, proper JSON output, and file attachment support. apfel represents a significant democratization of Apple's AI capabilities, allowing developers to leverage the Neural Engine and GPU for local inference without the costs or privacy concerns of cloud services. The MIT-licensed project demonstrates how Apple's hardware investments can be repurposed beyond their intended use cases.