Is a C API planned for Newton or Warp's runtime?

[ajaygunalan]

Newton's pipeline compiles Python functions to C++/CUDA code via Warp's JIT, which is then compiled to PTX and native libraries for GPU execution. Warp handles kernel compilation, GPU memory management, automatic differentiation, and spatial queries — all through a Python-only interface.

erwincoumans/warp_cpp demonstrates that pre-compiled Warp kernels can be called from C++ without Python at runtime, by loading cached compiled binaries. However this relies on undocumented internals.

Question: Is there any plan to formalize non-Python access into a stable C API at either layer?

1. Warp-level — stable interface for launching compiled kernels, managing GPU arrays, and using AD from non-Python code
2. Newton-level — C interface for Model creation, solver stepping, and state readback

This would follow the pattern of MuJoCo (C core → Python/Rust/Julia bindings) and PhysX (C++ core → C wrapper → Rust/C# bindings).

Context: The Rust robotics ecosystem has simulation (Rapier), visualization (Rerun), and linear algebra (nalgebra), but lacks a mature GPU-accelerated differentiable physics engine. Newton's solver stack fills that gap. A stable C API would unlock FFI bindings for Rust, C++, Julia, and others.

Has this been discussed?