gavinj.net

Thoughts on technology, infrastructure, and AI

Three Small ESP32 Devices, One Home Assistant

September 12, 2026 • home-assistant, esphome, esp32

My first three ESPHome projects turned a LILYGO T-Embed, a Nano ESP32 and a Waveshare reflective display into useful Home Assistant devices.

Eleven Servers, One Hammer, and a Serious Look at Proxmox

September 10, 2026 • proxmox, vmware, virtualization

A Broadcom upgrade conversation sent our infrastructure team looking at alternatives. Three retired servers, a graduate named Thor and a Proxmox trial have made the next renewal much more interesting.

One Jump Box, Many Users: Multi-User RDP on Ubuntu 26.04

September 3, 2026 • homelab, linux, ubuntu

Part 1 of building a hardened multi-user jump host: why the Essential Eight pushes you towards one, and how to get xrdp serving concurrent sessions on Wayland-native Ubuntu 26.04.

There's No Kimi Button: Custom Providers on Pangolin's New AI Gateway

August 30, 2026 • pangolin, ai-gateway, kimi

Pangolin 1.22 shipped an AI Gateway this week. The provider list covers OpenAI, Anthropic and Gemini, but there's no Kimi or DeepSeek button. Here's how to register both as custom providers, and why the missing buttons turn out not to matter.

The eggplant in the address bar: what an emoji subdomain taught me about the modern web

August 28, 2026 • emoji, dns, idn

My friend Rob registered 🍆.prawnburrito.com, pointed it at my blog, and gave me a reason to look at how emoji domains work, where they break, and why the standards bodies quietly wish people would stop.

gav.ilicio.us

Lifespan
After a few weeks listening to David Sinclair's podcast, I'm tempted to subscribe to Lifespan magazine. The future of longevity research is looking extraordinary.
Working with Git Worktrees in Magit
A practical Magit guide to Git worktrees—the neat trick for keeping parallel agent sessions isolated without cloning the whole repository.
Omarchy: Beautiful, fun & agentic Linux
Might be fun to spin this up!
mlc-ai/web-llm: High-performance In-browser LLM Inference Engine
Web-LLM is a high-performance in-browser LLM inference engine. Running models locally in the browser removes the need to send prompts to a server and opens up offline or privacy-sensitive use cases.
Reverse Engineering Unknown File Formats with ImHex
This article introduces ImHex by reverse engineering FEZ's save file format. ImHex has become one of the best hex editors I have found recently, and the pattern matching features are particularly useful for decomposing binary files.
'Just use a VPS bro' (OpenClaw version)
This video really encapsulates my life as a Linux admin at this point in time.
A Walkable ASCII Cyberpunk City in One HTML File
I think I'm going to buy this! A walkable ASCII cyberpunk city rendered in a single HTML file.
General Resolution: LLM usage in Debian
This was a really progressive resolution, you can't please everyone - the smartest people I know lean into this technology. I think this will ultimately accelerate Debian and FOSS in general.
OpenClaw 2.0, Accidentally
Bob might be getting an upgrade ...
binvis.io: Visual analysis of binary files in the browser
binvis.io is a browser-based tool for visualizing binary files. It maps bytes to colors using space-filling curves and entropy visualization, making it easier to spot structure, compressed sections, and encrypted regions without uploading data to a server.