About
You load a local model, it looks great, then you drop it into an agent loop and it quietly falls apart. A malformed tool call at step 7. A "task complete" when nothing got done. Public leaderboards don't test your GPU, your VRAM, or your quantization — the exact combination that breaks. QuantaMind runs the real agent loop on your own machine, fully offline, and gives a Ready / Conditional / Not Ready verdict per model and quant, with reasons named. It verifies tool calls actually happened instead of trusting the model's word, so scores can't be inflated. Unmeasurable on your backend? It says N/A, never a fake number. Works with Ollama, llama.cpp, and MLX. Fully open source, nothing leaves your machine. Feedback and contributions very welcome.
Badges
