Hugging Face releases AnyLanguageModel 1.0 for Apple platforms

Hugging Face released AnyLanguageModel 1.0 on October 20, marking a 1.0 milestone for its Swift framework spanning system, local and hosted language models. Developers work through one session interface across those deployment options.

Developers can replace Apple’s FoundationModels import with AnyLanguageModel and keep the same session code while choosing Apple’s system model, a local open-weight model or a hosted service. Hugging Face’s announcement names llama.cpp, MLX and Core ML as local backends.

For hosted inference, the framework supports OpenAI’s Chat Completions and Responses APIs, Open Responses-compatible endpoints, Anthropic’s Messages API, Google Gemini, Ollama, Private Cloud Compute on OS 27 and Hugging Face Inference Providers. A custom base URL also lets developers use other OpenAI-compatible services.

Hugging Face says guided generation, streaming tool calls and token-usage data work across local and hosted models. It also says every supported provider handles tool calling and guided generation, including structured generation for nested Generable types and arrays. The framework runs on Linux as well as Apple platforms.

The AnyLanguageModel 1.0 release notes require swift-huggingface 1.0, which in turn requires swift-xet 1.0. Hugging Face released both companion packages alongside AnyLanguageModel to support accelerated model downloads from the Hugging Face Hub in Swift apps.

The swift-huggingface 1.0 release settles that package’s public API, includes breaking type and method-name changes, and adds bucket file downloads with progress reporting. The swift-xet 1.0 release marks Xet.withDownloader, XetDownloader, its configuration and XetDownloaderError as stable public API.