Skip to content

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

[Unreleased — public-beta-1.0.0]

Planned

  • Scale to 40+ nodes, open access to all AI floor citizens
  • Bring-your-own-token access model with courtesy API tokens
  • Secrets management app (SOPS + GPG)
  • Indoor cabling and network hardware refresh

[Unreleased — alpha-0.1.0]

Planned

  • Acquire 4+ OptiPlex chargers (or USB-C with adapters) to power the flashed nodes
  • Bring 4 nodes online (1 CP + 3 workers)
  • Deploy NemoClaw instances for select floor members

Added

  • next-steps.md — planned work for alpha-0.1.0 and public-beta-1.0.0
  • Reorganized repository as a Read the Docs site (MkDocs + Material theme)
  • GitHub Pages deployment via GitHub Actions

[alpha-0.0.2] — 2026-03-23

Current release. The cluster is bootstrapped and operational on Talos Linux.

Added

  • runbook.md — complete step-by-step guide to bootstrapping the Talos Linux Kubernetes cluster (BIOS config, image building, flashing, bootstrapping, Cilium CNI installation)
  • inference-capacity.md — hardware tier comparison for LLM inference across 5 scenarios (OptiPlex CPU-only, eGPU, Mac Mini M4, multi-Mac, prosumer rig)
  • changelog.md

Changed

  • All docs migrated from k3s + Ubuntu plan to Talos Linux v1.12.5 + Kubernetes v1.35.2 + Cilium v1.19.1
  • plan.md — rewritten for Talos-based architecture (1 CP + 2 workers, 192.168.10.x network)
  • compute-capacity.md — updated for Talos overhead, current 3-node state with 20-node projections (renamed from CAPACITY.md)
  • gpu-inference.md — updated for Talos cluster (GPU nodes join via kubelet since Talos lacks NVIDIA driver support)
  • os-install.md — rewritten for Talos raw-image flashing (replaces Ubuntu manual/PXE strategy)

Contains

  • plan.md — cluster architecture, network topology, software stack, open questions
  • runbook.md — full bootstrapping and operations guide
  • compute-capacity.md — CPU, RAM, storage, workload capacity estimates
  • inference-capacity.md — LLM inference speed across hardware tiers
  • gpu-inference.md — Phase 2 GPU inference paths
  • os-install.md — Talos image flashing strategy

[alpha-0.0.1] — 2026-03-02

Initial release. Planning and capacity analysis for a 20-node k3s cluster on Ubuntu.

Added

  • README.md — project overview and contents index
  • plan.md — full cluster plan: hardware baseline, node roles, network topology, k3s software stack, installation phases, timeline
  • compute-capacity.md — cluster-wide compute estimates, LLM inference throughput, storage planning
  • gpu-inference.md — two paths to GPU inference (dedicated GPU nodes, eGPU via Thunderbolt 3)