@patmenciu

modelpilot runs repeatable two-round benchmarks on local ollama models to validate replacement decisions, measure speed and quality, verify no-think behavior, and audit installed models without uploading data or calling cloud apis.

qwen-nothink creates derived ollama tags that disable thinking output for local qwen models via manifest patching and chat template override, with cli and api verification steps included.

ollama-lifecycle provides structured workflows for model pulling, benchmarking, aliasing, reference scanning, and safe deletion. heavy on checklists and safety gates, but documentation leans toward procedural verbosity rather than sharp decision logic.