Spaces OS dispatch #6
Week 9 - Jul 03
"The purpose of abstraction is not to be vague, but to create a new semantic level in which one can be absolutely precise."
— Edsger W. Dijkstra
When you don't have a lot of updates, start with a quote. It makes it look like more is going on behind the scene.
Joke aside, we're completely re-architecturing how the agent, model and MCPs are communicating together to provide better security guarantees. We expect to have this land sometime next week.

For more details, take a look at the design document: Design: secure agent integrations
I expect some of that work to also be transferable to regular applications. The OS is not only about LLMs, but also to bring us to a more secure foundations. If we want to be able to execute applications generated by our friends and family, they also have to be reasonably sandboxed.
Audio support
Audio is still a topic. The nemotron model is showing good promise to give us live transcription support.
Once transcription is live, it creates new requirements. It becomes necessary to be able to backtrack. To remove filler words. Or to go back if the user says "actually". Words can also change based on context provided later in the sentence.
And when the model replies too fast, the user can feel rushed. Sometimes you need a moment to ponder and think about what you just said. Finding that balance is something that is going to take time.
Genode microvm progress
Our friends over at Qubes are still exploring the Genode as a better interface to run micro kernels on top. This is a long term project to give us a more secure foundation than the Linux kernel.
Now that virtiosfs (file sharing between the guest and the host) is working, the focus is now on virtio-gpu (graphic sharing); which includes some of the work previously done on Munix.

Next up
One this is done, we want to focus on more foundational work; application sandboxing, packaging and distribution.
Release
https://github.com/generational-infrastructure/spaces-os/releases/tag/spaces-2026.07.03