What I
use.

Updated 1 May 2026

Hardware, software, and what I deliberately dropped. No perfect stack, this is what works right now. The Spark is the hub; how I run local LLMs on it lives separately.

Lab · powered on

DGX Spark

Memory
128 GB unified
Compute
Grace Blackwell
Purpose
Local models + evals
Location
Hoorn
Models
Gemma-4 · Nemotron-3
On the go

MacBook Air M3

Memory
8 GB
OS
macOS
Use
Web work + writing
Vibe
Light and quiet
At home

Mac mini M2

Memory
16 GB
Screens
2× 4K
Keyboard
Apple Magic Keyboard
Mouse
Apple Magic Mouse
Desk
Sit-stand
Audio

For calls and recordings

Microphone
Rode PodMic
Setup
Desk mat + arm
Status
Works fine
Code

Editor & terminal

Editor
VS Code (→ Cursor todo)
Terminal
Apple Terminal
AI pair
Claude Code · Opus 4.8
Plan
Anthropic Max
Theme
Default, no time for it
AI / ML

Models & stack

Hosted
Anthropic Claude
Daily
ChatGPT for simple work
Local
Gemma-4-26-4b-it
Embeddings
Qwen-Embedding-4B
Search
PGVector
Tools
MCP servers where they fit
Development

Languages & frameworks

Backend
Python (pip + uv)
Frontend
TypeScript (npm)
Frameworks
Next.js · Vue · Astro
Data
Supabase · Prisma
React stack
Remix / RR v7
Namesake
Django (yes, literally too)
Infra

Hosting & ops

Apps
Vercel · Fly.io
Domains
TransIP
Edge
Vercel + Supabase
This blog
TransIP + GHA
Side bot
Pi 5 with OpenClaw
Dropped

Cloud GPUs for benchmarks

Too much noise between runs. One Spark, one thermal profile: reproducible numbers. Since the DGX moved in, I only run benchmarks there.

Dropped

Tooling chase

Switching editors, tweaking themes, maintaining dotfiles. Time I'd rather spend on evals. I only change a tool when it's genuinely in my way.

Dropped

Notion and heavy doc tools

Markdown in a repo does the same for me, faster, and without a cloud account that could change tomorrow.

Missed something or questions about a choice? Send a message.

Esc