dhh opened omacom/omarchy#16590 to add local coding models through the agents panel and command line. The first hardware-specific recipe targets NVIDIA N1X and GB10 machines with Qwen3.8 27B TensorFold and a 200k context. It would download pinned images and weights, run the server under rootless Podman and configure pi to use it.
The proposal remains open against Quattro. dhh reports a working add and restart flow on a 64 GB N1X laptop, but lists removal, model switching, fresh-machine setup and selection through the panel among the untested paths. Documentation and a dedicated icon are also still missing.
