AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms

Why it matters

AnyLanguageModel has reached its first stable release after a year of development, giving Apple developers a single Swift interface for both local and hosted language models. Built to mirror Apple's Foundation Models API, it lets apps swap the system model for open-weight alternatives running through llama.cpp, MLX, or Core ML, or for hosted services from OpenAI, Anthropic, Google, Ollama, and Hugging Face. Tool calling, guided generation, and image attachments work across providers, and the library also runs on Linux. It ships alongside swift-huggingface and swift-xet, which accelerate model downloads from the Hub. The project reflects a broader shift: small open-weight models have become capable enough to run on iPhones and Macs, while larger ones remain usable through hosted inference. The developers say the same approach will extend to additional model types, including models that decide and models that listen.