Apple’s macOS includes a privacy-focused AI language model, approximately 3 billion parameters in size, that developers can access through Apple’s Foundation Models framework. The on-device model operates entirely offline on Apple Intelligence-compatible devices, processing data locally to help ensure sensitive information remains on the Mac. According to Better Stack, this setup removes the need for internet connectivity, API keys, or external AI services. Developers can use an independent open-source tool called apfel, which offers three interfaces: a Command Line Interface (CLI), Chat Prompt Interface, and an OpenAI-compatible local server. The model supports tasks such as generating code snippets, summarizing text, rewriting content, and tagging data. Key advantages include local data processing for enhanced privacy, no cloud-related costs for on-device inference, and straightforward integration. Constraints include its smaller scale, limited context window, and Apple Intelligence compatibility requirements.