About
In short: Should We reads your bill and does the maths on your own device, in your browser. No AI works out the answer: a program replays every half hour of a year on every plan and battery size, the same way each time. What we keep is held in Sydney with anything that says who you are removed first.
Where it runs
- The site is a Next.js app, hosted on Vercel. Its server functions run in Sydney, so your data stays in Australia.
- No cloud keys are stored anywhere. Each copy of the site signs in to Amazon Web Services with a short-lived token that Vercel issues it, and the role it takes can only add files under its own folders. It can’t list or delete them.
- Every change to the code goes live on the next push.
The maths runs in your browser
- The engine is plain TypeScript with no network access. It replays a year of half-hourly use with a battery on each plan and each battery size, and reports the bill, the payback and the ten-year value for every combination.
- It runs in a background thread on your own device, so a usage file never has to leave your machine to get an answer.
- It is checked against an independent reference simulator written in Python, to one part in a million, and by more than 900 automated tests across the engine, the form, the bill readers and the chat tools.
Reading your bill
- Your bill’s text is read in the browser by a PDF library, not by an AI. The same bill reads the same way every time, and reading it costs nothing.
- The original file, and its name, never leave your device.
- The copy we keep is made on your device: the reader finds names, addresses, account and meter numbers, barcodes and QR codes, paints them black on a picture of each page, and builds a fresh PDF with the remaining text underneath. Nothing of the original file is carried over.
What is kept, and where
- A private, encrypted store in Sydney, run by Amazon Web Services, holds the blacked-out bill copies, the layouts of usage files we couldn’t read, each visit’s progress through the questions, and saved reports.
- A saved report is reached only by the random link in your browser’s address bar. Nothing can list them.
- The chat tools store nothing. A result’s link is its inputs and the engine’s version, encoded, and the same inputs always give the same answer, so a report is worked out again from its link rather than kept.
- Test copies of the site write to a separate folder, so testing never mixes with what real people sent.
- What is kept, for how long, and how to have it deleted is in our privacy policy.
Emails, such as a quote request, a report’s summary or a saved report’s link, are sent through Amazon’s email service in Sydney, from an address on our own domain. The email never leaves Australia.
In ChatGPT and Claude
- Should We is also a tool a chat can call. The same list of tools is served two ways: through the Model Context Protocol (MCP) at
shouldwe.com.au/mcpfor Claude and ChatGPT, and as a plain web API under/api/v1with an OpenAPI description built from the same list, so the two can’t drift apart. - The answer card you see inside the chat is the same card, drawn beside the reply in both ChatGPT and Claude.
- Each chat asks the same questions as the site, one at a time, and our engine does the maths. The chat reads the result; it doesn’t work out the figures itself.
- How to add it is on the chat page.
The figures behind the answer
- Synergy’s and Horizon Power’s plans, the federal and Western Australian battery rebates, the solar buyback schemes and the wholesale market’s prices are built into the site, each with its source, and shown when you hover a figure.
- Solar output for each weather point comes from the European Commission’s PVGIS solar data, prepared ahead of time.
- Every fact on a report shows its source. Every estimate and assumption says how it was worked out, and sits apart from the facts.
Questions
Should We is built by Luke Durrant in Perth. Questions to hello@shouldwe.com.au.