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

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

AnyLanguageModel is a new Swift package designed to streamline the integration of Large Language Models (LLMs) into Apple applications. It addresses the fragmentation of LLM APIs, allowing developers to use a single interface for both local models (Core ML, MLX) and remote providers (OpenAI, Anthropic).

Developers have historically faced complexity when integrating LLMs due to disparate APIs and requirements for local versus cloud-based models, leading to wasted time and discouraging experimentation with potentially efficient local solutions. This friction hinders the adoption of open-source models that could be ideal for many use cases.

Why This Matters

The current landscape of LLM integration for Apple platforms is fragmented, requiring developers to manage multiple APIs and integration patterns. This contrasts with the ideal of a unified, seamless experience. The cost of experimentation with different models is high, potentially costing developers significant time and resources before finding the optimal solution for their needs.

Key Insights

• Swift Package Traits: Enables developers to include only necessary dependencies, reducing bloat.
• Foundation Models as Base: Leveraging Apple’s API provides a familiar and stable foundation for LLM integration.
• Multi-Backend Support: AnyLanguageModel supports Apple Foundation Models, Core ML, MLX, llama.cpp, Ollama, OpenAI, Anthropic, Google Gemini, and Hugging Face Inference Providers.

Working Example

// Using Apple's built-in model
let model = SystemLanguageModel.default
let session = LanguageModelSession(model: model)
let response = try await session.respond(to: "Explain quantum computing in one sentence")
print(response.content)

// Using an open-source model running locally via MLX
let model = MLXLanguageModel(modelId: "mlx-community/Qwen3-4B-4bit")
let session = LanguageModelSession(model: model)
let response = try await session.respond(to: "Explain quantum computing in one sentence")
print(response.content)

Practical Applications

• Chat Applications: Building AI-powered chatbots with seamless switching between local and cloud models, like the chat-ui-swift example.
• Image Analysis: Extending applications to support vision-language models, despite limitations in Apple’s Foundation Models framework.

References:

• https://huggingface.co/blog/anylanguagemodel