Senior software engineer, 20+ years. Solo agency owner. Self-taught.
Going hard on local AI. 3x AMD AI Pro R9700's (96GB VRAM) on Arch Linux. Anti-hype, pro-receipts. If you can't run it on your own hardware, do you own it?
FasterChat - Blazing fast, privacy-first AI chat interface. Preact + Hono. Self-hostable. Open WebUI but actually open. ⭐ 152
ScaffoldBench - Opinionated local LLM benchmark. Tests real engineering tasks, not trivia questions. Because wall-clock time to a useful answer is what actually matters.
Scaffold Code - A framework for harnessing AI agents. The harness layer that makes agents do real work instead of vibes.
Claude Toolkit - My Claude Code configuration, skills, and workflow patterns. How I actually use agents day to day.
Knitly - Private, chronological community platform. No algorithms, no ads, no telemetry. Your people, your timeline, your space.
Claude Explorations - Persistent AI agent architecture. Memory graph in SQLite, Go maintenance pipeline, Signal integration. How I build agents that remember. ⭐ 3
Yeet - Minimal app launcher for Wayland. Rust. On the AUR. ⭐ 23
Homelab Homepage - Lightweight startpage for your homelab. System stats, calendar, service directory.
Arch Linux. AMD ROCm. llama.cpp. Preact. Hono. Go. SQLite. Caddy. Bun. If it's bloated, I probably ripped it out and replaced it with something lighter.
Self-hosted everything. I write about the homelab, local AI, and agent architecture on X and at mikekey.com.
- mikekey.com - A notebook since 2006
- @1337hero on X
- scaffoldbenchmark.com
I build infrastructure to protect what I value. FasterChat because your AI conversations shouldn't live on someone else's server. Knitly because your family's social graph shouldn't be owned by Meta. ScaffoldBench because the local model space needs honest benchmarks, not vibes.