Open GPU compute
for AI, science, and ships.
Navatala builds Apache-2.0 multi-backend GPU and AI libraries, then applies the same compute foundation to marine digital twins for ship design, inspection, operations, and compliance.
$ pip install navatala-gpu
>>> import navatala_gpu
>>> navatala_gpu.__version__
'0.1.1'
>>> navatala_gpu.get_capabilities()
{'extension_loaded': True}Developer preview
2,080 CUDA kernels, generated across five GPU backends.
CUDA and HIP/ROCm currently have the broadest generated coverage. OpenCL, Vulkan, and Metal are included with explicit backend limitations.
Open compute, proven in maritime engineering
Open GPU + AI Libraries
- Multi-backend runtime for CUDA, HIP/ROCm, OpenCL, Vulkan, and Metal
- AI/ML kernels for clustering, KNN, regression, trees, vector search, DNN primitives, and dataframe operations
- Scientific-computing kernels for BLAS-like operations, sparse solvers, reductions, scans, and CFD primitives
- Public coverage matrix and PyPI package for early adopters
Marine Digital Twin
- Hull, hydrostatics, stability, and structural design
- Configurable inspection workflows and evidence capture
- Voyage, fuel, emissions, EU FuelEU, and IMO DCS workflows
- GPU-accelerated CFD and analytics as the compute-heavy layer matures
Why Navatala
Specification-driven
Formal contracts keep APIs, generated kernels, runtime behaviour, and application workflows consistent as the system grows across platforms and domains.
Cross-platform by construction
GPU kernels target CUDA, HIP/ROCm, OpenCL, Vulkan compute, and Metal; application code targets desktop, mobile, web, and server deployments.
Offline-first
Digital-twin apps work without connectivity. Delta sync ensures nothing is lost when you come back online.
Portable GPU + AI kernels
An open kernel corpus covering AI/ML primitives, vector search, dataframe operations, sparse solvers, reductions, scans, and CFD building blocks.
Regulatory compliance
Built-in support for EU FuelEU Maritime, IMO DCS, Carbon Intensity Indicator, and classification-society requirements in the operational twin.
Open and honest
Apache-2.0 release of the GPU and AI kernels. Public coverage matrix per backend. Developer-preview status stated up front, not buried in footnotes.
Build with the library, or build the vessel twin.
Use the open GPU package, or talk to us about applying the same compute foundation to marine design and operations.