by Optimal Intellect
Moreau
The Control Layer for AI
Input
AI Model
Control Layer
Moreau
optimal · constraints enforced
∇ differentiable
Output
Real System
AI predicts well. But real systems have limits.
Moreau guarantees every output is safe and optimal.
The Problem
AI doesn't understand constraints
Robots must obey physics
Joint limits, torque bounds, contact dynamics
Portfolios must respect risk limits
Leverage caps, sector bounds, regulatory constraints
Grids must balance supply and demand
Capacity limits, power flow physics, safety margins
Without a control layer,
AI cannot be deployed in the real world.
The Solution
Why Moreau?
A differentiable optimization layer that guarantees every output satisfies your constraints—while enabling end-to-end learning.
Differentiable
End-to-end learning with hard guarantees. Backprop through the control layer.
Batched
Run the control layer with large batches—128 to 1024 problems at once.
GPU-Native
No CPU bottleneck. All computation stays in VRAM throughout training.
AI-Compatible
PyTorch and JAX native. Drops into your existing ML pipeline.
Performance
CPU is already fast
GPU is even faster
benchmarked on NVIDIA H100
4-14×
faster
Power Systems
9-35×
faster
Model Predictive Control
25-99×
faster
Portfolio Optimization
What industry are you in?
How We Work
We work with you
Moreau isn't a black box you download. We partner with you to integrate control layers into your system.
Design with you
Understand your constraints and model your problem correctly.
Build with you
Integrate Moreau into your existing ML stack.
Support you
Enterprise-grade support as your needs evolve.
Team
Built by optimization experts
The team behind CVXPY, CVXPYlayers, and 50+ research papers on convex optimization.
Creators of CVXPY (3M+ downloads/mo)
Creators of CVXPYlayers (900+ citations)
Stanford PhDs from Boyd's lab
50+ papers on optimization