11 products.
3 layers.
1 self-improving system.
An existing product family built on the SpinForge core. Explore differentiable optimization, scientific evaluation, autonomous discovery, algorithm improvement and the Recursive Engine that connects the family.
Find the part
that serves your work.
Five foundational products
SpinForge Warp
Differentiable annealing that lets a learning workflow backpropagate through optimization.
AMDE + ALCHEMI
Physics-grounded drug-discovery scoring through GPU molecular dynamics.
Neural Schedule
Learned temperature schedules for annealing through a Fourier neural operator.
GARMO Scientist
An autonomous discovery product organized around perceive → hypothesize → experiment → learn.
Annealed Splatting
Hybrid 3D reconstruction using annealing for structure and gradients for parameters.
Five connected applications and optimizers
QC Bridge
A quantum–classical optimization bridge for warm starts, validation and hybrid workflows.
Gödel Optimizer
A self-evolving optimizer that improves its optimization algorithms.
Materials Factory
A self-improving materials-data pipeline built on the SpinForge core.
Embodied Optimizer
Physical optimization in which a robot solves combinatorial problems.
Token Factory
GPU-fleet scheduling formulated as a QUBO to maximize tokens per watt.
The cross-product meta-system
The eleven products are real products in the owner-supplied README table. SpinForge Warp and Embodied Optimizer are included; names from a different architecture diagram are not substituted. “Self-improving” is the family’s stated architecture, not a universal performance rate.
A shared engine.
Distinct discovery outcomes.
Bench has no asserted Discovery Core dependency. Recursive Engine is the synthesis meta-system, not another name for this engine.
Existing products.
New connections to qualify.
Checked improvement feedback
The existing Recursive Engine and Gödel Optimizer inform the connection; cross-family integrations still have explicit acceptance boundaries.
RESEARCHThe geometry of answers
Understand the analysis role separately from candidate scores.
THE FULL INVENTORY64 named entries
Explore every application, engine and module without counting wrappers twice.