Dubnium — Summary

Overview

Dubnium is a single-host NixOS configuration for the workstation target. WSL is a headless validation target for shared modules and docs, not the deployed workstation. Mode changes use an explicit desired-state / observed-state model and a reconciler that enforces mode via systemd targets, services, and slices.

Choose your path

Install workstation

Start with Fresh Install and Custom Installer USB.

Bring up existing host

Use First Bring-Up, then transition validation.

Enable local inference

Seed models with Model Seeding, then opt into vLLM Runtime.

Work on memory

Start with Persistent Memory and Memory Service.

Current defaults

dubnium.vllm.enable = false — vLLM is opt-in for explicit compute testing.
dubnium.plano.enable = false — Plano routing is opt-in until its runtime is installed and validated.
Runtime and user secrets stay outside Nix source. See Runtime Secrets.
User-level Home Manager configuration comes from external/dotfiles on ryjen/dotfiles feat/nix-migration.

Install source contract

Installer media labels are DUB-ISO and DUB-SEED.
Install bootstrap uses local source from media or checkout state, not an install-time GitHub token.
Post-install source reconciliation is explicit. See Post-Install Source Reconciliation.

Ownership boundaries

Owner	Responsibility
Dubnium	NixOS system config, workstation services, install media, runtime units
Dotfiles	Home Manager user config, user shell, user-level tool configuration
Runtime secret provider	Host and user secrets outside the Git-tracked Nix source
Model/router repos	Client policy, routing schemas, and model-router behavior

Modes

desktop — interactive default, GUI on, local AI off (v1)
studio-local — overlay on desktop for audio-priority policy; AI suppressed
compute — headless, graphical session absent, vLLM/AI active

Architecture highlights

Mode changes are requested explicitly and verified against the host's observed state.
Desktop, studio, platform, and AI workloads have separate operating policies.
Transitions are guarded so the machine does not silently enter a mixed or unsafe mode.
The design can adapt as workstation hardware changes, including GPU topology.

Purpose

Provide a reproducible, policy-driven way to move a single host between interactive and compute-oriented roles safely and observably. Useful for a workstation that doubles as a local AI node and low-latency audio studio.

Dotfiles

Dubnium owns system-level configuration. User-level configuration is owned by ryjen/dotfiles feat/nix-migration, checked out locally at external/dotfiles. See External Sources.

vLLM

vLLM is a compute workload, but it is disabled by default on the workstation. Enable it only for explicit compute testing after model seeding and runtime setup. Avoid broad PyTorch, audio, JAX, or TPU extras unless they are explicitly required.

Access

The summarized site is hosted for private consumption and is accessible over the micrantha VPN for authorized users. Contact @micrantha to request access.