fe72581c · zzmq.dev@gmail.com/zz
Commit
fe72581c2d3e4fa0307b62e4af4ac821a0bf4f7b
Model selection, per-turn timing visibility, lower default max-turns zz work --model <name> picks any locally available Ollama model (zz models lists them with size/parameter count, current default marked). Addresses "I want to select my model from all the ones available" directly. Each turn now logs "waiting for <model>..." before the request and "got a response in Ns" after -- a non-streaming call to a 14B-class local model can genuinely take 10-30+ seconds with nothing printed in that window, which reads as hung rather than working. This makes the actual per-turn cost visible and comparable instead of opaque. Lowered the default max-turns per claimed ask from 200 to 60: every turn resends the whole growing conversation, so an unbounded attempt gets slower the longer it runs, not just longer in wall-clock terms. 60 is still well above what any observed real completion has needed, and isActuallyDone's nudging already ends a stuck attempt earlier in the common case -- this is just the hard backstop, now tighter. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Troy Howard <thoward37@gmail.com> -- 2026-10-10T12:06:55.000Z