AnyLanguageModel 1.0 unifies access to local and hosted models

AnyLanguageModel 1.0 has been released, providing access to local and hosted models through a common API for Swift. It builds on the Foundation Models interface: after changing the import, the same session code can be used with the system model, local models through llama.cpp, MLX and Core ML, and hosted services.

According to the authors, all providers support tool calling, including streaming output, and structured generation, including nested @Generable types and arrays of those types. The library also allows images to be attached to prompts using Foundation Models 27 syntax; for the system model, this capability is available on iOS, macOS and visionOS version 27.

Supported services include APIs from OpenAI, Anthropic and Google, Hugging Face Inference Providers, and other compatible interfaces. The library also runs on Linux, allowing the same code to run on a server. It now retrieves API keys through a function instead of a fixed string, allowing the application to fetch short-lived tokens from its own server.

The choice of local model remains dependent on hardware. The authors cite Qwen3.8 27B for MLX as an example: its 4-bit variant takes up approximately 16 GB and, according to the authors, runs on a Mac with at least 32 GB of memory. For models too large for a given device, Hugging Face Inference Providers can be used with the same session code.