Superstructure
Results before you commit
Change a load, a section, a material — or move a node — and see what it does before you decide to keep it. Analysis stops being a step you run and becomes part of drawing.
Free for non-commercial use
Understand the structure. Right now.
Results that keep up with your edits. A whole team in one model. Every change traced, comparable and reversible.
Placeholder · editing a frame with results updating live
Engineers already know what a structure should do. What they lack is a tool that lets them see it — continuously, while they work. Without that, judgement gets hedged. The hedges compound, and they end up in the steel.
Superstructure is built around three things that have to be true at once.
A team in one model
Several engineers in the same model at the same time, each running their own analysis. An AI agent can be one of them, sent in from a meeting. An admin view of who is where, and review before work lands.
Every change accounted for
Split, compare and bring structural models back together the way developers do code. Options sit side by side. Change control and replay come as standard.
Contingency is the hidden cost
Every design carries margins that have nothing to do with code factors: a load allowance added just in case, an unwritten 80% utilisation ceiling, deflection criteria never pushed to the edge, elements grouped by broad result ranges, a buffer for the architectural change everyone knows is coming.
Each one is reasonable on its own. Stacked, they are expensive.
- Moynihan and Allwood, examining 23 steel-framed buildings at Cambridge, found roughly 36% of the steel could have been removed without breaching code compliance.
- Cambridge’s MEICON survey measured average steel utilisation at 0.56 against a 0.8 design target — 20–50% more material than necessary.
- Structure accounts for 50–80% of a building’s upfront embodied carbon, which is why this is now a regulatory question as much as a commercial one.
This is not bad engineering. It is what happens when the tool cannot answer what is governing here? quickly enough to be worth asking. Make that question cheap, and the contingencies become a decision instead of a habit.
Get out of the way
The model is the thing worth looking at. Chrome retreats, results appear where the question was asked, and an applied view stays put while you keep working.
A file on your machine
The model is one file on your machine, with its whole history inside. It opens without a server, without an account and without a connection. The analysis runs on the same machine, so the results you see were computed from your own copy.
A team shares a model through one small server. It puts everyone’s edits in one order and holds the shared model; the analysis still runs on each engineer’s machine. The server we run today is one machine in the EU, not a US cloud.
Try it
There is a working build. It is published nightly, runs on Windows, macOS and Linux, and connects to a shared demo model you can edit alongside anyone else who is in it. You can also just watch that model being edited, live.