Small, sharp tools that end the game — against slow local LLMs, bad configs, and everyday computer nonsense. Everything measured on real hardware, tested hard, and quiet by design: no telemetry, no accounts, no nonsense.
Benchmarks your models on your actual GPU: finds the largest context that stays 100% on GPU, catches silent truncation and CPU spill, and applies the optimal config. The full report is free; $49 for one-click apply, KV-cache A/B, and the dashboard.
The Windows 11 ghost-popup fix — for that maddening invisible rectangle that eats clicks over Explorer. Free.
A lean Rust proxy that gives coding agents proper control over local Qwen models — including the think switch the OpenAI-compat endpoint ignores. Free.