Privacy Policy
Last updated 8 October 2026
Your blueprints are commercially sensitive. This page says exactly what we collect, who else touches it, how long we keep it, and how to get it back or have it destroyed — without the hedging that usually fills a page like this.
Pending legal review. This text describes how the product actually works and is written to be relied on, but it has not yet been reviewed by a lawyer. If anything here matters to a decision you are making, ask us and we will answer plainly.
The short version
- We collect what is needed to run your account and read your drawings. Nothing else.
- Your uploaded documents are never used to train AI models. Not ours, not anyone’s.
- We do not learn from your corrections either. A fix an estimator makes stays in your project and changes nothing about how the software reads anybody else’s drawings.
- Your company’s data is isolated from every other company’s in the database itself, not just in the interface.
- You can export everything at any time, and delete everything permanently.
- We do not sell data, and we do not run advertising trackers.
What we collect
Account information. Your name, email address, and company name. If you sign in with Google, we receive your name, email and profile picture from Google — not your password.
The drawings you upload. PDF plan sets and the pages within them, plus anything the system reads from them: sheet numbers and titles, room names, dimensions, schedules, legends and notes.
The work you produce. Takeoffs, quantities, pricing, estimates and proposals, including edits you make and who made them. Review history is deliberate: the point of the product is being able to show where a number came from and who approved it.
Wall suggestions you turn down. On plans with wall suggestions (beta), a suggestion an estimator rejects instead of confirming is stored with your project, so it is not offered again on another device, and a wall deleted from your takeoff is kept under Removed so it can be put back. It is your company’s data like the rest of your work, visible only to your company, and it does not change how the software reads anyone’s drawings.
Usage and billing records. Which features are used, how many sheets were analysed, projects consumed, and subscription status. Payment card details are handled by Stripe and never reach our servers.
Acceptance records. When you accept our terms we record your name as typed, the date, the version of the text, and the IP address the acceptance came from. That record exists so both sides can establish what was agreed and when.
Support messages. Whatever you send us, so we can answer it.
Takeoff checklist requests. If you ask for the takeoff checklist, we keep your email address, when you asked, the page you asked from, your answers to its twelve questions (so we can see which gaps are common), and whether you ticked the separate box to receive the occasional email about interior estimating, with the words that box showed. The address is used to send you the checklist and nothing else unless you ticked that box, and the record of your choice is kept so we can show what you agreed to. It is kept apart from accounts and is not added to any other list. Every occasional email carries a link to stop them, and you can ask us to delete the record at any time.
How we improve accuracy, and what we use to do it
Your drawings are never used to train AI models. Not ours, and not our providers’. Uploaded documents are sent to Anthropic for the sole purpose of reading them and returning results to you, and are not retained for training. That is a commitment we hold them to as well as ourselves.
We do not currently learn from your corrections either. When an estimator fixes something the software read wrongly, that fix stays in your project and goes nowhere else. No record of it is kept outside your own company, and nothing about how the software reads drawings changes because of it. As of 9 September 2026 not one correction has ever been recorded.
We are saying this because an earlier version of this page said the opposite — that corrections were collected and used to improve detection for everyone. That was written alongside the database table built to hold them, before anything was connected to it, and it stayed on the page after it stopped being true. It was never true in practice: the table has always been empty.
If that ever changes, you will be told before it does. The table is deliberately narrow if it is ever switched on — the database itself permits only the field corrected, the value the software produced, and the value the estimator replaced it with, limited to units, CSI classification and quantities. Item descriptions are excluded outright, because a description reads “interior door 3070, room 104” and that is about your job. Values are capped in length and cannot be structured data, so a record physically could not hold a plan set, a price breakdown, or a client’s details. None of that is running today.
No other customer ever sees your work. Your projects, drawings, quantities and proposals are visible only to your own company. Nothing you do in the product reaches another account, and no competitor gains anything from your plan set.
We state this plainly rather than burying it because “we may use your data to improve our services” is the sentence that hides everything. It hides both directions: what a company is taking, and what it only says it is taking.
Who else processes your data
We use a small number of sub-processors. Each is here because the product cannot work without it.
- Supabase — database, sign-in, and file storage. Your drawings live here, and the email confirming your address is sent from here.
- Anthropic — the AI that reads the drawings. Documents are sent for processing and returned; they are not retained for training.
- Vercel — hosting, and the cookieless count of visits to the public pages. Handles web requests; does not store your documents.
- Stripe — subscription billing. Holds your payment details. We never see a full card number.
- Intuit QuickBooks — only if your company connects QuickBooks Online. When you press Send to QuickBooks, that proposal’s customer name, lines, prices and tax go to your QuickBooks company as an estimate. Nothing is sent otherwise, and disconnecting deletes the connection here.
- Resend — sends the emails the product sends you: confirming your address, resetting a password, a team invitation, the guides you ask for, and notices about your account such as the date it will be deleted. It handles your email address and the message, and nothing else.
- Google Workspace — our own mailbox, which receives your replies and carries anything we write to you by hand.
- Google Ads — measurement for the ads we run, and only on the pages outside your account. It is told that a signup happened and which ad preceded it. It is never told who you are, and it is not present while you work.
We will update this list before adding anyone new to it, not after.
How your data is kept separate
Every row of your data carries your company’s identifier, and the database refuses to return a row belonging to another company — regardless of what the application asks for. This is enforced by the database, not by the interface, so a bug in a screen cannot expose another company’s work.
We test this rather than assert it: an automated check runs against every table, proving one company cannot read or write another’s data, and a companion check deliberately breaks the protection to prove the test would notice.
How long we keep things
Deleted projects: 30 days. A deleted project sits in the trash and can be restored. At 30 days it is permanently removed, including the uploaded files — except a project with money on record (an accepted proposal or a recorded payment), which is kept, in the trash, so you can always see where that money came from. The Trash says which projects are kept and why. Closing the account removes those too.
Closed accounts: 30 days. Closing your account starts the same 30-day clock. During it, everything can be restored exactly as it was. At the end, all of it is permanently deleted — projects, drawings, takeoffs, estimates, proposals, and the files in storage.
Ended subscriptions: 90 days, read-only. When a subscription ends, the account stays for 90 days: everything can be opened and exported, and nothing changed. We email the owner at least 14 days before the end, with the date. Subscribing again restores it as it was; otherwise it is then permanently deleted, as a closed account is.
Free accounts nobody uses: drawings after 12 months. If nobody in a free account (one with no subscription) opens the app for 12 months, the drawing files uploaded to it are deleted; its projects, takeoffs, estimates and proposals are kept. We email the owner at least 30 days before, with the date, and signing in before then keeps everything.
Exporting does not stop deletion. It means you hold your own copy. We will tell you the exact date your data will be destroyed, not just “in 30 days”.
Billing and tax records are kept longer where the law requires it. These are invoices and payment records, not your drawings.
Getting your data out, or destroyed
Export. Settings → Export your data has all of it in one place: each project’s drawings as uploaded, its takeoff and estimate as Excel or CSV, and its proposals as PDF, or several projects at once as one .zip, projects in the trash included. Your original uploaded files remain downloadable for as long as the project exists, except a free account’s after 12 months without use (above).
Deletion. Delete individual projects from the project list, or close the whole account from Settings. Both are permanent once the 30 days elapse, except a project with money on record, which is kept until the account is closed (above).
Access and correction. Ask us and we will tell you what we hold about you, correct it, or delete it. Under Canadian privacy law (PIPEDA) you have the right to both; we would do it if you asked regardless.
Cookies
We use cookies for four things. The first is keeping you signed in: a session cookie set when you sign in and cleared when you sign out.
The second only applies if you arrive through a referral partner’s link (an address like preconstructai.com/r/…). That link sets one cookie on this site holding the partner’s code — nothing about you — and when you opened it, for up to 30 days, so that if your company signs up, the partner who introduced you is credited. When the link is opened we record the visit for that partner: the page it came from and the type of browser, but not your IP address. If your company signs up, the partner can see its name; they never see anything inside your account.
The third is advertising measurement, and it exists only on the pages outside your account. When we advertise on Google, our marketing pages and the signup page load Google’s tag, which sets a cookie recording which ad brought you here, so we can tell whether an ad led to anyone actually signing up. It is not loaded anywhere inside your account. Once you are signed in and working on a bid, no part of the page reports to Google — the single exception is the “set up your workspace” screen a new account sees once, which is where a signup is counted.
We never send Google your email address or your name. Google’s “enhanced conversions” setting — the feature that uploads customer emails — is switched off. What is sent when an account is created is the fact that one was, and a one-way hash we cannot reverse and Google cannot match to anything, which exists so that one signup is not counted twice.
The fourth is how we know which site sent you. If you arrive from somewhere else — a post on LinkedIn, a search result, a link a colleague sent — one cookie records the name of that site and nothing else, for up to 30 days, so that if your company later opens an account we can tell what brought you. We keep the name of the site, never the address of the page. A web address can carry a search you typed or the name of a document, so only the site’s name — linkedin.com, google.com — is ever stored, along with the campaign tag on the link if it had one. Moving between pages of this site is not a source and is not recorded.
It holds nothing about you: no name, no address, no identifier, and nothing that follows you to another site. If your company opens an account, the name of that site is saved against the company and the cookie is deleted. If you never sign up, it expires on its own. We do this because we are a new company with no customers yet, and knowing which of the things we write is worth writing is the difference between finding the next fifty and guessing.
We do count visits, and we do it without a cookie. On those same pages outside your account — never inside it — page views are counted by Vercel’s own analytics. Vercel already hosts this site and is already on the list above, so measuring the pages adds nobody new to the people who see anything about you. It sets no cookie and issues no identifier, which means it cannot follow you to another site or recognise you here tomorrow. What it records is the page, the site that linked to you, a country, and whether you are on a phone or a desktop. It is there so we can tell whether anything we wrote was read at all.
There is no session recording and no heatmap tool, and nothing measures you inside the app. Once you are signed in and working on a bid, no part of the page reports anywhere — not to Google, not to Vercel, not to us. The one exception is the signup count described above, on the “set up your workspace” screen a new account sees once. If you would rather not be measured by the advertising cookie on the way in, blocking third-party scripts in your browser, or Google’s own ad settings, stops it; nothing on this site works any differently without it.
Security
Data is encrypted in transit and at rest. Access to your files requires a signed, time-limited link tied to your account. Administrative access is restricted to a single named account, and every administrative action is recorded in an audit log the person taking the action cannot edit.
See the Security page for detail.
Where your data is held
Our database and file storage are hosted by Supabase, and hosting by Vercel. Data may be processed in the United States as well as Canada. AI processing happens on Anthropic’s infrastructure, which is US-based.
If your work is subject to a requirement that data stay within Canada, tell us before uploading anything, because at present we cannot guarantee it.
Children
This is a tool for construction professionals. It is not intended for anyone under 18, and we do not knowingly collect their information.
Changes to this policy
We may update this policy. The current version is always on this page, with the date it was last updated at the top. A material change takes effect when it is posted here, and the next time you open the app it is shown to you in a popup that says what changed, which you confirm you have seen before carrying on.
Questions about anything on this page? Email luca@preconstructai.com or use the contact form, and we will answer in plain language.
PreConstruct AI