Install the Desktop Application

🚧

Tensormorph has not shipped a public release yet, so specific download links, installer packages, and version numbers aren't available. This page describes what installing will involve based on confirmed platform and backend support, not a finished install flow.

Supported platforms

Tensormorph's compute layer targets five backends - CPU, wgpu/WebGPU, CUDA, ROCm/HIP, and Metal - which implies desktop support across Windows, Linux, and macOS, with GPU acceleration available wherever a matching backend (CUDA on NVIDIA, ROCm on AMD, Metal on Apple Silicon) is present. See GPU Backend Compatibility for how backend selection works once installed - you are not required to have a GPU at all, since CPU is a first-class backend, not a fallback.

What to expect

Based on the application's design, installing and first-running Tensormorph will involve:

Downloading a platform-appropriate installer or package.
A first-launch step where Tensormorph detects available compute backends on your machine and sets a sensible Compute: Auto default (see GPU Backend Compatibility).
An empty workbench, ready for you to open a checkpoint - see Open Your First Model.

Command-line tooling

Beyond the desktop application, Tensormorph ships CLI tooling (tmorph) for indexing and validating checkpoints outside an interactive session - see Tensormorph CLI - which is installed separately and doesn't require the desktop app to run.

Related resources

Open Your First Model - the next step once installed.
GPU Backend Compatibility - what determines which backend Tensormorph uses on your machine.
Settings - where your compute and appearance preferences live once set.

What is Tensormorph?Open your first .safetensors model