DEVOPSdigest

Qt Group is launching Qt AI Assistant, an experimental tool for streamlining cross-platform user interface (UI) development.

Its goal is to reduce repetitive UI development tasks that often do not involve coding. This allows developers to spend more time on productive and creative aspects of writing innovative code. The tool supports self-hosted language models (via cloud or locally) without the need for a third-party AI assistance provider.

Qt AI Assistant will offer advice on building application UIs with Qt Framework and automates manual tasks like writing unit test cases, code documentation, and repetitive code. Users request this advice with a prompt window and quick-access commands inside the code editor of Qt Creator (used for creating and modifying cross-platform apps).

The tool is compatible with various coding languages like C++ and Python, but specifically trained on thousands of real-world use cases of QML and Qt Quick – the Qt Framework’s language and toolkit for designing how apps look and behave. This means developers can, for example, generate the needed repetitive QML code while building user interfaces, freeing up time for more complex coding tasks.

Qt AI Assistant will accommodate whichever language model developers prefer to deploy with. The initial release is pre-configured to support the following models out of the box:

- Claude 3.5 Sonnet (chat + completions)

- OpenAI GPT-4o (chat + completions)

- Llama 3.3 70B QML (chat)

- Code Llama 13B QML (code completions)

- Code Llama 7B (code completions)

"We’ve been hearing from developers that they’re increasingly spending time on tedious tasks that don’t involve coding. Qt AI Assistant is part of our efforts to correct that. We want to eliminate the chores in developers’ day-to-day, because software development should be about writing great code,” says Peter Schneider, Senior Product Lead at Qt Group. “The industry has been buzzing with AI announcements – everyone’s got their favourite language model, and we don’t want to take that away from anyone. It was important that we support an open approach and allow enterprises to freely choose their preferred LLM deployment method.”

The flexibility of Qt AI Assistant will allow for different content generation requests to be routed separately to different models. For example, a developer can route QML code to one specialised LLM, while routing other code to another LLM. This lets developers always obtain code suggestions from the best-performing models for their programming language, instead of whichever LLM is available through a third-party AI assistant provider.

Qt AI Assistant will also address security concerns for embedded devices and high-end desktop applications built with Qt Framework. By supporting self-hosted models, companies can perform private cloud deployments of LLMs, which avoids code leaks and safeguards intellectual property. Developers can even choose to only trigger Qt AI Assistant manually, to avoid unwanted disruptions while coding.

"Our mission is to help customers increase productivity throughout the product development lifecycle," says Juhapekka Niemi, Senior Vice President, Product Management, at Qt Group. "Qt AI Assistant is an important step towards our vision of enabling rapid UI development and reducing project timelines from months to weeks, and eventually days."

Later this year, Qt Group will release fine-tuned versions of LLMs – starting with Llama 3.3 70B and Code Llama 13B – trained on 4,000+ human-created and validated QML code snippets. These will be free to download on the Hugging Face portal, including for open-source Qt developers to connect with third-party assistants.