Your AI, tried a million times.

Gnomon plugs into the AI you already use. Before your AI hands you code, a design or a decision, Gnomon runs it up to a million ways: code on up to a million inputs, a design in a million mixes of colour, type and spacing, a decision through a million futures. Your AI comes back with the version that holds up, and how many of the million passed.

No. Gnomon is not an app you open. You add it to your AI once and keep working the way you do today. Your AI checks its work with Gnomon whenever it matters, without you asking.

For code and design, Claude Code: Gnomon runs on your computer and checks every change. For decisions, Claude, ChatGPT and Grok: you add Gnomon as a connector. Developers can also call it from the Claude, OpenAI and xAI APIs or their own software.

When an AI coding tool like Claude Code changes a function, Gnomon runs the old version and the new one on the same inputs, 100,000 by default and up to a million, leaning on awkward cases like zero, empty text and very large numbers. Every input where the answers differ is shown with both results, and Claude must explain or fix it. On a copy of JENA's code it caught 3 of 3 planted bugs; JENA's own tests caught 2.

It tries a million combinations of colours, text sizes, line heights and widths, keeps your brand colours, and returns the ones that pass WCAG contrast and readability rules as ready-to-use CSS. It also opens your pages in Chrome at many screen widths and text sizes to find overlapping or cut-off text, sideways scrolling and buttons too small to tap.

No. The code and design checks run on your own computer, next to your code. Functions that might reach networks, files, databases, email or payments are listed, never run, and keys and passwords are removed from the environment the code runs in.

Gnomon tries many versions of the same thing and counts how many work. For code, that is up to a million inputs; for a design, a million mixes; for a decision, a million futures with the uncertain numbers, like next month's sales, changed each time. That last method is called a Monte Carlo simulation.

Very fast. A million simulations of a quote take about 70 milliseconds, and a whole year of cash flow under half a second, so the check fits inside a normal AI reply. A code change is checked on 100,000 inputs in a few seconds, and a million design mixes take about a second.

No tool can promise that. Agreeing on 100,000 inputs is strong evidence, not proof, and a simulation shows how likely an outcome is, not a promise. What Gnomon does is catch the mistakes your AI would otherwise hand you, and give you a count to decide with.

Code and design checks never leave your computer. For decisions sent to the Gnomon API, what you send is checked in memory and dropped when the answer is sent; Gnomon keeps only an id, the time, your key, sizes, the result and timing for each call.

No. The checks run your code, measure your design and simulate your decision without any AI model, so the same question always gives the same answer.

Gnomon is in beta. A test key is free with 100 checks a day, and beta partners get up to 100,000 units a month free. After launch the API price is NZ$0.03 per 1,000 units; a million-simulation check is 4 units, about a hundredth of a cent.

For decisions in the Claude app: open Settings, go to Connectors, add a custom connector and paste https://gnomon-api.deskwithnova.com/mcp/demo. No sign-in is needed for the demo. For code and design in Claude Code: install the Gnomon package (beta partners get access) and run claude mcp add gnomon-local -- gnomon-local.

No-mon. A gnomon is the part of a sundial that casts the shadow showing the true time.

Gnomon is powered by NOVA Enterprise Limited, a New Zealand AI company that also makes JENA, Mini NOVA and NOVA Ink.