Spaces OS dispatch #4

Week 7 - Jun 19th

New forge

In the spirit of keeping everything self-hosted and dogfooding our own Clan deployment tool, the code is now available on Tangled, a new P2P Git forge based on ATProto. We're keeping the GitHub mirror for back-compat.

https://tangled.org/spacesos.org/spaces-os

Voice integration

Kenji has been busy fixing the voice integration. We now have a proper recording notifier to make it more transparent when voice-to-text is running.

Harness rework

The system grew from a desktop-only, to server and web agents. It was time to revisit how we are putting the agents together, and to unify them all behind a single daemon (pi-sessiond).

Desktop agent

Something that was missed from previous comms, you can now use Command+L to launch new agents. It's convenient if you want to just launch a quick task in the background.

The new agent then appears in the desktop bar, pulsing until the work is done:

Web agent

We spent some time styling the web agent.

Misc fixes

New models

This week two new interesting models releases stand out:

  1. https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b is a better voice-to-text model we're exsited to try out next week.
  2. GLM 5.2 was released, an open model that outperforms GPT 5.5 in some benchmarks. https://artificialanalysis.ai/articles/glm-5-2-is-the-new-leading-open-weights-model-on-the-artificial-analysis-intelligence-index