apfel exposes the built-in Apple Silicon FoundationModels LLM as a UNIX CLI tool and a local OpenAI-compatible server, enabling on-device inference with no API keys, cloud, or network requirements.
Project overview
It bridges a practical gap for macOS 26 Tahoe+ users by turning a closed on-device model into standard developer interfaces, making fully private local LLM workflows viable without any paid services.
Project type
Model Runtime
Use cases
Chat Assistants · Coding & Development
Deployment
Refer to project documentation
License
MIT
Best for
Developers and general users on Apple Silicon looking for a fully local, private LLM interface for UNIX workflows and existing OpenAI-compatible tools.
Key capabilities
A pipe-friendly CLI for single prompts, piping input, file attachments, and generating output.
A local server providing a drop-in http://localhost:11434/v1 backend.
An interactive REPL for testing prompts or MCP servers.
Attach Model Context Protocol servers via --mcp flag; apfel discovers, invokes, and returns results.
Attach files, PDFs, or images directly to prompts for on-device text extraction, OCR, and image understanding.
Guaranteed schema-valid JSON output via guided generation.
Native ToolDefinition and JSON detection support for tool calling in all contexts.
Real shell scripts wrapping apfel (cmd, oneliner, mac-narrator, etc.) bundled in the binary.
One model (~3B params on-device), not configurable.
Apple's safety system may block benign prompts.
Getting started
Install easily via Homebrew without Xcode, then run a prompt directly: `brew install apfel` followed by `apfel "What is the capital of Austria?"`.
Evidence and sources
README: Apple Silicon Macs ship a built-in LLM via [Apple FoundationModels](https://developer.apple.com/documentation/foundationmodels). `apfel` exposes it as a UNIX tool and a local Open…
README: `apfel --chat` is a small REPL for testing prompts or MCP servers. For a GUI chat app, see [apfel-chat](https://github.com/Arthur-Ficial/apfel-chat).