A Coding Agent on Every Desk, Part 3: A Fast Model in a Sidecar Software AI Home Lab DIY Last time in this series, I talked about configuring my home lab to run on two GPUs, which itself followed part one of configuring the home lab. This time I didn’t add a second GPU,b ut a second local model: a small, fast “classifier” model next to the main server, doing the small, fast tasks the large model is the wrong fit for.
A Coding Agent on Every Desk, Part 2: Two GPUs and a Bigger Context Window Software AI Home Lab DIY The second graphics card arrived, and so did a new model. This is a follow-up to A Coding Agent on Every Desk. The Podman, Tailscale, and Open WebUI pieces from that post are unchanged. What changed is the card in the second slot, the model behind the qwen-coder alias, and most of the llama-server command line.
Rethinking Build vs. Rent in the Era of Coding Agents Software AI Infrastructure Libraries Testing Back in 2014, I wrote a little thought experiment called Not Invented Here Mechanic.
