Ubuntu integrates local AI

Jon Seager, VP of Engineering at Canonical, explained in his post The future of AI in Ubuntu how Canonical intends to integrate AI into upcoming Ubuntu versions. Locally installed open-weight models are to be used to improve existing functions such as speech-to-text for accessibility and to introduce new AI functions such as assistance with errors and automation through agentic workflows. Context-sensitive help, for example with troubleshooting or complex tasks, is also planned.

The current LTS version Ubuntu 26.04 already facilitates the use of local AI: Developers can install the AI stacks for Nvidia and AMD hardware (CUDA and ROCm) with a simple apt command from the standard package sources. In addition, there are the Inference Snaps, with which a number of AI models can be easily installed locally, including the multimodal AI model Qwen VL, the reasoning model DeepSeek R1, the open Google model Gemma 3, and the Nvidia models Nemotron 3 Nano and Nemotron 3 Nano Omni. The system is intended to install the desired model in a variant and quantization that suits the user's hardware. Canonical is working with CPU and GPU manufacturers for this purpose. Like all Snaps, the AI models run in a sandbox with restricted access to the system.

Positive reactions from the community

Community reactions to Seager's post are overwhelmingly positive. Ubuntu users praise the measured use of AI and the refusal to force AI functions on users. The use of local models instead of large AI models from the cloud is welcomed, even if only small, less powerful models run on many notebooks and PCs.

In a later post, Seager clarifies that all integrated AI functions in Ubuntu are to be set up as uninstallable Snaps – users who do not want to use them can simply get rid of them. Ubuntu 26.10 will feature an initial preview of the AI functions as opt-in, meaning users must explicitly agree. Later, the question of AI usage could be integrated into the installer. The intention is not to force AI on anyone. Seager reiterated that the plans exclusively involve the use of locally installed language models.

(odi)