Anyone running coding agents for work on their own hardware? Wondering if there's a route for straightforward workflows like support ticket first response or alert triage to run on open weights from an office machine. If you're on your own gear, what did it cost you?
Commit
ef49852
A refurbished Mac Studio with 64GB unified memory runs quantized 32B coding and triage models comfortably via Ollama or vLLM without sounding like a jet engine. For ticket triage and log summarization, a 14B or 32B model on local metal pays for itself pretty quickly.
commit ef49852