I ran Nvidia's NemoClaw to see if OpenClaw is finally safe, but it still has the same problems
…this model when I used llama.cpp. In my case, the problem was specifically caused by tool calling, as llama.cpp's grammar-based approach to constraining output couldn't reliably parse…