Starting with macOS Tahoe, Apple Silicon Macs will include a built-in on-device large language model as part of Apple Intelligence. This model runs entirely locally using the Neural Engine and GPU, ensuring no data is sent to the cloud and no API keys are required. While Apple exposes this capability through the FoundationModels Swift framework, it is currently limited to system features like Siri and Writing Tools. There is no direct command-line interface or public HTTP endpoint available for general developers to interact with the model out of the box. A new open-source project called apfel aims to bridge this gap. This Swift 6.3 binary tool wraps the underlying LanguageModelSession to make the native model accessible for broader development. apfel transforms the restricted system feature into a versatile utility that functions as a UNIX command-line tool, an OpenAI-compatible HTTP server, and an interactive chat application with context management. As a command-line interface, apfel allows users to pipe text into the model via standard input and output, providing a method to interact with the AI without building a full application. For developers who prefer an API approach, apfel runs an HTTP server built on the Hummingbird framework, mimicking the standard OpenAI API structure. This allows existing tools and scripts designed for OpenAI to work seamlessly with Apple's local model. The tool addresses several limitations inherent in Apple's raw API. It provides proper exit codes for better error handling, supports JSON output for programmatic integration, and enables file attachments. Recognizing that the model operates with a relatively small 4096-token window, apfel implements five different context trimming strategies to manage memory and conversation flow effectively. Additionally, it offers real token counting through the SDK and automatically converts OpenAI tool schemas into Apple's native Transcript.ToolDefinition format, simplifying the integration of external tools. By leveraging the power of the on-device model, apfel demonstrates that Apple Intelligence can be used outside of the Apple ecosystem's walled garden. It enables developers to utilize the privacy-focused, local AI capabilities directly on their Macs without reliance on external servers. This approach maintains the security and speed advantages of on-device processing while unlocking the potential for custom applications and automated workflows. The release of apfel represents a significant step for the community in unlocking the full potential of Apple's new machine learning infrastructure.