Updates on Oznog

What's Next

Fri, 16 Jan 2026 00:00:00 +0000

Node0 is operational.

The DI fleet runs inference. Storage serves data. NixOS keeps configs sane. The core works.

But “operational” isn’t “finished.” Here’s what’s next.

Immediate: Services #

The hardware runs. Now it needs to do more.

Ollama/vLLM deployment. Proper inference servers with API endpoints. Not just interactive sessions. Services that other services can call.

n8n workflows. Automation connecting the pieces. DI capabilities wired into daily workflows. The infrastructure doing work while I do other things.

The Collaboration Model

Thu, 15 Jan 2026 00:00:00 +0000

Every post in this series mentions collaboration with Claude.

That’s not marketing. It’s how the work actually happened.

What Collaboration Means #

Not “AI wrote it for me.” Not “I told AI what to do.”

Something in between. Something harder to describe.

Planning together. “Here’s what I’m thinking for the ZFS dataset structure. What am I missing?” Back and forth until the design is solid.

What Would Be Different

Wed, 14 Jan 2026 00:00:00 +0000

Twenty-two months from hardware purchase to operational fleet.

Not the fastest path. Not the slowest. The path that happened.

Looking back: what would I do differently?

Start with Storage #

This one I got right.

nexus went up first. ZFS. Centralized datasets. NFS exports ready before any compute came online.

Every machine that followed mounted storage from day one. No local data silos. No migration headaches later. The pattern was established early.

The DI Fleet: 128GB+ VRAM, Running Independently

Tue, 13 Jan 2026 00:00:00 +0000

Four DI servers. All operational.

128GB+ of VRAM across the fleet. The ability to run 70B parameter models locally. Inference, fine-tuning, training. All on infrastructure I control.

Full local capability. Complete control. The freedom to experiment with whatever the work requires.

This is what we built.

Node0 v1 deployment. The DI fleet running: di01-di04 visible with their color-coded fans, nexus storage at bottom, UPS systems providing protection. This is the initial deployment - two 42U racks with the full hardware buildout are still to come. But this is operational now.

di01: The Heavy Iron

Mon, 12 Jan 2026 00:00:00 +0000

The utility tier was up. The production server was stable. Time for the flagship.

di01. The heavy iron. The machine for when you need everything.

The Specs #

Component	Spec
CPU	Threadripper Pro 5975WX (32 cores)
RAM	512GB ECC DDR4
GPU	2x RTX 4090 (watercooled)
VRAM	48GB total
Storage	NVMe + NFS from nexus
Fans	Red (Simon Says)
Power	~350W idle, dedicated UPS

32 cores. 512GB ECC. Dual watercooled 4090s. This is not a subtle machine.

di02: Production Server

Sun, 11 Jan 2026 00:00:00 +0000

di03 and di04 proved the patterns. Now it was time for the production tier.

di02. The workhorse. Stable inference endpoints that don’t go down for experiments.

The Specs #

Component	Spec
CPU	Threadripper Pro 5945WX (12 cores)
RAM	448GB ECC DDR4
GPU	2x RTX 3090 (NVLink bridge)
VRAM	48GB total
Storage	mdadm RAID 1 (OS) + NFS from nexus
Fans	Yellow (Simon Says)

Threadripper Pro for ECC memory. 448GB of it. When production inference serves real workloads, memory errors aren’t acceptable.

di03 and di04: The Utility Servers

Wed, 07 Jan 2026 00:00:00 +0000

Not every workload needs a Threadripper.

Before deploying the heavy iron, I brought up the utility tier. di03 and di04. Smaller machines. Essential roles.

di03: Utility (Capable) #

Component	Spec
CPU	Ryzen 9 5950X (16 cores)
RAM	128GB DDR4
GPU	RTX 4090 (AIO watercooled)
VRAM	24GB
Storage	NVMe + NFS from nexus
Fans	Green (Simon Says)

The “capable” utility server. 14B-32B models run comfortably. Heavier batch processing. Backup inference capacity when the big machines are busy.

Fiber: Real Bandwidth

Sat, 03 Jan 2026 00:00:00 +0000

The storage foundation was in place. The network had a name. But the pipe to the outside world was still local point-to-point wireless internet. Good company. I’ve been grateful for them, but it’s affected by weather and is limited to 150mbps.

Time to fix that.

The Upgrade #

Allwest Fiber. 2.5 Gbps down, 1.6 Gbps up. Symmetric-ish. Real bandwidth.

PTP wireless was fine for consumption. Streaming, browsing, the usual. But sovereign infrastructure needs to push as much as it pulls. Backups to offsite. Syncing between locations. Serving content. Hosting services.

NixOS Standardization

Fri, 02 Jan 2026 00:00:00 +0000

The mental shift to config as code on the Framework was practice.

Now it was time for the real thing. NixOS across the fleet.

The Problem #

Every Linux machine is a snowflake. Install packages manually. Configure services by hand. Document what you did (maybe). Hope you remember it later (probably not).

Three dev boxes. Four DI servers coming. A storage server. Utility machines. Each one configured differently. Each one a liability.

Node0 Is Real

Wed, 31 Dec 2025 00:00:00 +0000

December 31, 2025. Last day of the year.

The five-day sprint was behind us. Storage deployed. Services running. But something was off.

The network was still called TempleOfHarmony.

A Name From Another Life #

TempleOfHarmony made sense once. A different phase. A different purpose.

But this infrastructure wasn’t a temple. It was a node. The first node of something larger.

Names matter. They shape how you think about things. Every time I saw “templeofharmony” in a hostname, it felt wrong. A past identity on present infrastructure.

Five Days of Building

Tue, 30 Dec 2025 00:00:00 +0000

December 27 through 30, 2025. Five days. One sprint. A sovereign computing foundation that actually works.

This is what got built.

Day 1: Storage Goes Live #

The storage foundation came first. Nexus, the Terramaster F8-424 Pro, went from box to production.

Created ZFS pool with RAIDZ2 (two-drive fault tolerance)
Configured datasets with appropriate recordsizes
Migrated ~21 TB from old Synology boxes
Deployed Restic REST server for encrypted backups

By evening, 21 terabytes of data had a new home. The Synology boxes could finally retire.

The Storage Foundation

Sat, 27 Dec 2025 00:00:00 +0000

After months of the hardware waiting, it was finally time to build.

December 27, 2025. The Terramaster F8-424 Pro, codename “nexus”, went from box to production in a single day.

The Iron #

Component	Spec
Chassis	Terramaster F8-424 Pro
Drives	8x Corsair MP600 PRO NH 8TB NVMe
Pool	ZFS RAIDZ2
Usable	41.3 TiB
RAM	48 GB
Network	10GbE

Eight NVMe drives. Not spinning rust. Not SATA. Full NVMe speed across the entire pool.

The Future We Were Promised

Thu, 25 Dec 2025 00:00:00 +0000

Christmas morning. A package I’d been waiting for since pre-ordering months ago.

Inside: a Commodore 64 Ultimate. The first real Commodore computer in over 30 years. Not an emulator. Not software pretending. An FPGA recreation of the original silicon, wrapped in a clear case with blue LEDs.

And a card with a sticker:

Technology That Enables #

In 1982, Commodore launched what would become the best-selling desktop computer of all time. Not because it was the most powerful. Because it invited you in.

The Tools That Stuck

Tue, 25 Nov 2025 00:00:00 +0000

A month into the Framework 13 experiment. Time to take stock.

Which of the modern CLI tools actually stuck? Which ones am I reaching for daily?

The Survivors #

These earned permanent spots in muscle memory:

fish for interactive shell. The autocompletions alone justify it. Tab through command history with context. Syntax highlighting as you type. Sane defaults without configuration sprawl.

Config as Code: The Mental Shift

Sat, 22 Nov 2025 00:00:00 +0000

Two days into the Framework 13 setup, something clicked.

I wasn’t just learning new tools. I was learning a different way of thinking about configuration.

The Old Way #

Twenty years of muscle memory: write scripts. Bash, PowerShell, whatever. Step-by-step instructions. Do this, then this, then this.

# The imperative approach
apt update
apt install nginx
systemctl enable nginx
systemctl start nginx
echo "server { ... }" > /etc/nginx/nginx.conf
systemctl reload nginx

It works. Until it doesn’t. Until you need to run it again on a different machine and something’s different. Until you need to know what state the system is in rather than what commands were run.

Framework 13: Sovereign Laptop

Thu, 20 Nov 2025 00:00:00 +0000

The hardware had waited for months. But before deploying servers, I needed a development environment that could drive the work.

November 2025. Framework 13 with Omarchy. The laptop that would orchestrate everything.

The Machine #

Component	Spec
CPU	AMD Ryzen AI 9 HX 370
RAM	96GB DDR5
Storage	2.8TB NVMe (btrfs)
OS	Omarchy (Arch + Hyprland)

Framework because repairable, upgradeable, not locked down. Started with 64GB RAM, upgraded to 96GB when the work demanded it. That’s the point. Hardware you can modify.

The Hardware That Waited

Wed, 20 Mar 2024 00:00:00 +0000

This post is dated March 2024-when the hardware was purchased. But it’s written in late 2025, looking back with the perspective that only comes after the waiting is over and the building has begun.

In March 2024, something shifted.

I don’t remember an aha moment. That’s not how I work. When something hits and I know it, I go all in. Within weeks I’d spent a substantial amount (like stupid substantial) on hardware. Servers, switches, storage.