Search tldr

apfel - Your Mac Already Has AI

apfel exposes Apple’s built-in on-device language model as a zero-configuration CLI, local OpenAI-compatible server, and chat tool. It targets Apple Silicon Macs on macOS Tahoe, emphasizing privacy, scripting, and no token costs.

Share

apfel - Your Mac Already Has AI

Author: Arthur-Ficial | Published: Unknown | Generated: 2026-04-03 | Domain: apfel.franzai.com
Tags: ‘#local-ai’ ‘#macos’ ‘#apple-intelligence’ ‘#cli’ ‘#mcp’ ‘#openai-compatible’


TLDR

apfel is an open-source Swift 6.3 tool that turns Apple’s Foundation Models framework into a UNIX-friendly CLI, interactive chat client, and OpenAI-compatible local server. It uses the roughly 3B-parameter Apple Foundation Model already installed with Apple Intelligence on Apple Silicon Macs running macOS Tahoe, avoiding model downloads, API keys, cloud calls, and token fees. Its 4,096-token context window makes it best suited to shell automation, short text transformations, summarization, translation, and lightweight tool use rather than long-context or complex reasoning workloads.

Key Takeaways

  • Zero-config local inference: Install with brew install apfel; the tool uses the system model supplied through Apple’s FoundationModels framework, requiring Apple Silicon, macOS 26 Tahoe or newer, and Apple Intelligence enabled.
  • Three access modes: apfel provides a composable CLI with stdin/stdout and JSON output, --chat for managed multi-turn sessions, and --serve for an OpenAI-compatible API at http://127.0.0.1:11434.
  • Native MCP tool calling: --mcp connects local or remote Model Context Protocol servers, auto-discovers tools, executes calls, and feeds results back to the model across CLI, chat, and server modes.
  • Small fixed model constraints: Apple’s model is described as roughly 3B parameters, using mixed 2/4-bit quantization with a combined 4,096-token input/output context window; it is useful for concise transformations and scripts but limited for factual recall, math, large documents, and complex code generation.
  • Privacy and interoperability focus: The project claims no telemetry, analytics, background network calls, or data collection; its OpenAI API support includes streaming, tool calling, GET /v1/models, JSON response formats, CORS, and common generation parameters.

Images & Media

Keep reading