bonni

Run `npx skills add ymahlau/bonni --all -g -y` in your terminal to install all skills in this repository globally.

It trains neural network surrogates on both function values and gradients, so the model learns the landscape much faster and needs fewer expensive evaluations.

It optimizes any black-box function that returns a value and gradient, such as nanophotonics inverse design, Bragg reflectors, and grating couplers.

Yes. It ships a CLAUDE.md orchestrator and a plan-then-implement skill that guide Claude Code through structured feature development in this codebase.

No. The scipy-style optimize_bonni API only requires your function, its bounds, and a few settings to start optimizing.