During the concluding moments of the WWDC25 keynote last Monday, Craig Federighi announced that Apple had “expanded” its concept for Swift Assist, enabling developers to incorporate their preferred Large Language Model directly into Xcode.
We now have additional insights into how this functionality will be implemented.
What is Swift Assist?
This feature was introduced at last year’s WWDC but has only recently been launched with the Xcode 26 beta. Initially, Apple described it as:
“a companion for all of a developer’s coding tasks, allowing them to concentrate on higher-level problems and solutions. It is seamlessly integrated into Xcode, aware of the latest software development kits (SDKs) and Swift language features, ensuring developers receive the latest code functions that perfectly fit their projects. With Swift Assist, tasks such as exploring new frameworks and trying out fresh ideas are merely a request away. Swift Assist employs a robust model that operates in the cloud — and consistent with all Apple developer services, it is designed with privacy and security in mind. The code of developers is only utilized for processing requests and is never retained on servers, with Apple refraining from using it for training machine learning models.”
What is the new Swift Assist?
Apple has now elaborated on its novel vision for Swift Assist in Xcode 26, which features a built-in model for predictive code completion, a newly added native ChatGPT integration with limited free usage, and an avenue for developers to incorporate third-party providers (including local models running on their own devices).
By default, developers can enable ChatGPT with just a few clicks, subject to a daily request limitation. Those with a ChatGPT Plus subscription can log in or enter their API key to exceed these limits.
The true strength lies in the newfound versatility of Xcode: developers can add additional providers, like Anthropic, by simply entering an API key, granting access to the latest iterations of industry-leading frontier models for coding. Furthermore, Apple’s implementation enables developers to select which models to showcase from each provider and mark favorites for easier access.
Local models are also supported. If developers utilize tools such as Ollama or LM Studio, Xcode allows interaction with models running directly on their Mac. They have the option to add multiple providers and switch between them using Xcode’s Coding Assistant when starting a new conversation.
This initiative aligns Xcode with the evolving landscape of AI tools: model-agnostic, customizable, and modular, independent of Apple’s rollout of updated model versions. This is undoubtedly a positive direction, offering the flexibility that could enhance Xcode’s appeal in a multi-model environment.
Are you looking forward to using LLMs in Xcode? Have you been utilizing LLMs for your coding needs? Share your thoughts in the comments.
