Introducing AnyLanguageModel: One API for Local and Remote LLMs on Apple Platforms

Source: HuggingFace Date: 2025-11-20 URL: https://huggingface.co/blog/anylanguagemodel

Summary

Library release: AnyLanguageModel, a Swift package providing a unified API for LLM providers on Apple platforms (macOS, iOS). Backends: Apple Foundation Models, Core ML, MLX, llama.cpp, Ollama, OpenAI, Anthropic, Google Gemini, and HF Inference Providers. Swift 6.1 package traits for optional dependencies. Core API stable, pre-1.0. Roadmap: tool calling, MCP integration, guided generation.

Implications

Thread: open-weights ecosystem health. AnyLanguageModel on Apple platforms addresses a real fragmentation problem: Apple Silicon developers currently have 5+ incompatible local inference stacks (Core ML, MLX, llama.cpp, Ollama, Foundation Models) with no common abstraction. A single import-and-swap API dramatically lowers the cost of experimenting with different backends in production apps. The inclusion of Foundation Models (macOS 26+) alongside open-weight backends is the critical design choice — it means the same app code can use on-device Apple intelligence or a local llama.cpp model with a one-line change. Roadmap MCP integration is the one to watch.