- Apfel is an open-source tool that exposes Apple’s built-in on-device language model — normally locked behind Siri — as a CLI tool, an interactive chat, and an OpenAI-compatible HTTP server.
- The project has gained over 1,300 GitHub stars and reached 670 points on Hacker News, reflecting strong developer interest in accessing Apple Silicon’s local AI capabilities.
- Apfel supports the Model Context Protocol (MCP), enabling Apple’s on-device model to use external tools such as APIs, databases, and math services.
- The tool requires Apple Silicon, macOS Tahoe (macOS 26), and Apple Intelligence enabled on the device.
What Happened
A developer using the handle Arthur-Ficial released Apfel, an open-source Swift application that unlocks Apple’s built-in on-device language model for direct use outside of Siri. The project was posted as a Show HN on Hacker News on April 3, 2026, where it accumulated 670 points. The tool provides three interfaces to Apple’s SystemLanguageModel: a UNIX command-line tool with stdin/stdout support, an OpenAI-compatible HTTP server, and an interactive chat mode.
Why It Matters
Starting with macOS 26 (Tahoe), every Apple Silicon Mac ships with a language model as part of Apple Intelligence. Apple exposes this model through the FoundationModels framework — a Swift API that gives apps access to the SystemLanguageModel. All inference runs locally on the Neural Engine and GPU, with no network calls, no cloud dependency, and no API keys required. However, Apple provides no terminal command, HTTP endpoint, or way to pipe text through the model without writing a Swift application from scratch.
Apfel fills this gap by wrapping the FoundationModels API in a developer-friendly package. The project is significant because it transforms a locked-down system feature into a general-purpose local AI tool, allowing developers to integrate Apple’s on-device model into shell scripts, web applications, and automation workflows without any cloud costs.
Technical Details
Apfel is built with Swift 6.3 and wraps Apple’s LanguageModelSession API. It exposes the on-device model in three modes: as a UNIX CLI tool with proper exit codes, JSON output, and file attachment support; as an HTTP server built on the Hummingbird framework that implements the OpenAI API specification at localhost:11434; and as an interactive chat with automatic context management.
The tool handles capabilities that Apple’s raw API does not provide out of the box, including five context trimming strategies for the model’s 4,096-token context window, real token counting via the SDK, and conversion of OpenAI tool schemas to Apple’s native Transcript.ToolDefinition format. The OpenAI-compatible server supports streaming, tool calling, CORS headers, and structured response formats, allowing any OpenAI SDK client to connect to the local model by changing only the endpoint URL.
Apfel also implements the Model Context Protocol (MCP), which allows the on-device model to use external tools. By pointing Apfel at any MCP server, users can give the local model capabilities such as math computation, API access, database queries, and other functions that the base model cannot perform alone. The project includes demo shell scripts for tasks like natural language to shell command conversion, git commit summarization, and codebase orientation.
Who’s Affected
Mac developers and power users with Apple Silicon machines running macOS Tahoe are the primary audience. The tool is particularly relevant for developers who want local AI inference without cloud API costs, privacy-conscious users who prefer on-device processing, and automation engineers who want to integrate LLM capabilities into shell scripts and CI/CD pipelines. The OpenAI-compatible server also makes Apfel useful for developers who want to test applications locally before deploying them against cloud-based models.
Apple’s developer relations team may take note of the project, as it demonstrates strong demand for direct access to the on-device model that Apple has so far restricted to Siri and Writing Tools. The project’s rapid GitHub adoption suggests developers want more flexible access to Apple Intelligence capabilities.
What’s Next
Apfel is installable via Homebrew with the command brew install Arthur-Ficial/tap/apfel. The project’s GitHub repository continues to receive contributions, and the developer community response suggests demand for additional features and platform support. Whether Apple will respond by offering its own official CLI or API access to the on-device model — or take steps to restrict third-party wrappers — remains an open question for the macOS developer ecosystem.
