RightNow AI
July 19, 2025 — Update
Training AGI at the Metal Layer
AGI will not come from prompt engineering or running code in notebooks. It will not run through cloud APIs. Real AGI will be trained directly on hardware, and right now nobody is building the tools for that.
So I decided to build them:D
I started RightNow with a single GTX 1060 because I wanted full control. NVIDIA GPUs are very powerful, but their real potential is blocked by slow and heavy software. Tools like NVIDIA Nsight are powerful, but they are slow to use and built for generating reports, not for developers who want to move fast and build real systems.
RightNow is a GPU-native development environment built for speed. It has real-time CUDA profiling, inline kernel stats, multi-GPU orchestration, and hardware-aware insights without extra layers or delays. It is made so code and hardware work together as efficiently as possible.
Today's large language models cannot write complex system-level code well enough for this work. But with the right environment, the right tools, and full control over hardware, they will be able to.
I have gone through four pivots to reach this point. Each pivot removed distractions and focused on what matters. If AGI happens, it will not be built on another chatbot or demo. It will be built on a stack designed for hardware, from the kernel up.
This is what that stack looks like:
We are building the missing layer that lets developers control how AI uses the hardware.
"My mission is to build the low-level GPU infrastructure AGI will need before AGI exists."
I built this solo on a single GTX 1060 in Jordan. Now I'm ready to scale with the right partners.
If you want to be part of this, contact me.
