Tensor Network Editor

Draw tensor networks, save reusable JSON designs, and generate Python for supported backends.
Keep notes, groups, metadata, colors, tensor values, and reusable subnetworks inside the same saved design.
Export the current design as PNG, SVG, PDF, TikZ/LaTeX, Graphviz/DOT, or Mermaid, or keep working from generated code outside the editor.

Start from an empty canvas, built-in or session templates, a saved JSON design, or imported generated Python.
Use Templates and Template settings to choose graph size, boundary options, and other template parameters before inserting.
Arrange tensors on the canvas, then add metadata, notes, and tensor initialization only where they help your model.
Use Ctrl/Cmd + wheel to zoom, drag empty canvas space to pan, and use the minimap to re-center quickly.

Subnetwork library lets you browse reusable fragments, insert them later with fresh ids, or copy them into the current session templates.
Auto layout in Reflow can tidy the current selection or the whole graph after imports and edits.
Tensor Initialization lets you keep generated zeros, ones, fill values, or explicit JSON values per tensor.
Metadata filters highlight matching tensors, indices, groups, notes, or hyperedges without hiding anything.

Use the planner to build a manual contraction path and compare it with Auto full, Auto future, and Auto past.
Benchmark mode compares saved manual schemes against the automatic variants for the current design.
Preview and comparison rows help you see when an automatic proposal improves or worsens cost metrics before accepting it.

Hyperedges are available in normal mode and behave like first-class graph elements.
For unidimensional, For bidimensional, and For Tree help you model periodic families when you need them.
Planner editing and Benchmark mode are not available for designs with hyperedges, and advanced planner workflows remain centered on the normal mode flow.

Name	FLOP	MAC	Peak	Peak Memory