Privacy policy
What we store, who processes it, how long we keep it, and how to have it deleted. Last updated 1 November 2026.
HVAC Builder is an engineering tool. Most of what it holds is equipment data you typed — coil geometries, weather years, system graphs. This page is about the rest.
Who we are
HVAC Builder is run by Myron Van Staden, a sole trader established in Ireland who trades as HVAC Builder; his address is in the terms of service. He is the controller of the personal data this page describes: write to privacy@hvacbuilder.app about any of it. The supervisory authority where we are established is Ireland's Data Protection Commission. The rest of our agreement with you is in the terms of service.
What we collect
| Data | Why we have it |
|---|---|
| Account identifier — the subject id issued by our sign-in provider, plus your plan, token balance and usage counters | To attach your work to your account and meter what your plan includes |
| Your designs — saved systems, buildings, simulation runs, and the weather years you import or generate | They are the product; they are yours — listed for your account (or your workspace) and nobody else's |
| Supplier profile and listings, if you register as a supplier — company, contact email, territory reps, published components and systems, product links and images | Published listings are public by design; that is what a directory is |
| Quote requests (RFQ) — the contact email and message of whoever asks a supplier for a quote; what it asks about — an item, or a summary of the design, and the whole design only when its sender ticks the box to share it; the quantity and region, if given; and, for a sender who is signed in, the account identifier, which the supplier sees too | To pass the enquiry to the supplier it was addressed to |
| Contact form messages — your message and what it is about, the email address you give, your name and the address of a page if you give them, and the language of the page you wrote from | To answer you, and to act on a report of illegal content |
| Credentials — SHA-256 hashes of API keys and OAuth tokens (never the values), registered OAuth clients, and your own AI provider key if you store one | To authenticate programmatic access you asked for |
| Billing identifiers — the customer and subscription ids issued by our payment processor, the billing email given at checkout, and the date a cancelled plan ends | To reconcile a subscription with an account, and to confirm a subscription or its cancellation by email. We never see or store card details. |
We do not run advertising or analytics trackers. There is no third-party analytics script on any page of this site, and no advertising cookies. The only cookies in play are the ones our sign-in provider sets to keep you signed in. We do count page views on the public site — one number per day, page path and language, kept for 400 days — and nothing about the visitor: no cookie, no IP address, no referrer, no record of the request. The browser's user-agent string is read to leave search-engine and advertising crawlers out of the count — and by the rate limit below, to give a crawler that names itself its own allowance — and is not kept by either; the web server's access log, below, is the one place it is written down. We count what the app does the same way — solves, simulations, saved designs, quote requests, connector calls, assistant messages, first sign-ins, and whether each solve worked or which kind of error stopped it — one number per day, kind and plan, with nothing about who did it. A supplier sees the same kind of count for its own listings — how often each was solved with, simulated, reported on and saved — never by whom, and never the design. When a visit arrives on one of our own advertising links, we count it under that campaign's name — the name we wrote into the link — and read nothing else from the address.
Rate limits. To stop the catalog being harvested in bulk, the server counts how many answers carrying catalog figures have gone to a network in the last 5 minutes: a catalog or refrigerant page, the compressor lookup, the compressor picker in the app, and a directory or company page asked to read its machines at a condition of your choosing. What it holds while it counts is the network the request came from — the first two or three numbers of the address, never the address itself — and a count. All of it is in memory only: a count lapses with its 5-minute window; the list our operator watches a harvest on keeps a network — its first two numbers, how many requests, when the first and the last — for 60 minutes after its last request; nothing is written to disk or to the page-view figures. To judge a request it reads two things more and keeps neither itself: the user-agent string, and whether the page that linked to it was one of ours. Two doors that take no account are counted the same way, by network and in memory: a connector registering itself and a connector starting an authorization. Signed-out quote requests, and messages through the contact form, are limited by address and email, for an hour. None of it is linked to an account, a session or a page you read.
Access log. The web server in front of the app (Caddy) writes one line to a file on our host for each request to hvacbuilder.app — pages and the app's own calls alike (the staging site is not logged): the time, the site name, the method, the page path with the query string cut off, the response status and size, how long it took, and the browser's user-agent string. It does not write the IP address (or the port), the referrer, any cookie or other header, or what is after the question mark in an address. Its one use is to tell a crawler or a script from a visitor and to see what is loading the server; the operator reads it by hand, and it is linked to no account, session or person. The file closes when it reaches 5 MiB — about a day at ordinary traffic, sooner when a crawler is busy — and a closed file is deleted 7 days later; at most fourteen closed files are kept. Nothing from it goes to the page-view figures, the rate limit or any third party.
How it is used
To run the product you asked for: solving and simulating your designs, storing them under your account, metering your plan, billing your subscription, publishing the listings you choose to publish, delivering quote requests to the supplier they name, answering what you write to us, and acting on a report of illegal content as the terms describe. Nothing else. We do not sell personal data, and we do not use your designs to train AI models.
Who processes it
These are the only third parties involved: the ones our servers talk to on your behalf, the one that holds our mailboxes, and the one your browser may be sent to.
| Processor | What reaches them |
|---|---|
| Clerk — authentication | Your sign-in identity. They hold your email and password/social identity; we hold only the identifier they issue. |
| Stripe — payments | Your payment details, which go to Stripe directly and never through our servers. |
| Resend — email delivery | The contents of a quote request, emailed to the supplier it was sent to; the contents of a bug report or feedback you file in the app, emailed to us; a message sent through the contact form, emailed to us; a supplier's claim on a maker's page — the company, its account and its contact address — emailed to us, because we confirm each claim by hand; and the notices we send you: a subscription's or a top-up's confirmation, and product notes if you asked for them. |
| Google — our mailboxes (Google Workspace) | Every email sent to leads@, bugs@ or privacy@hvacbuilder.app and every email we send from them — our answers, the confirmations section 19 of the terms asks for, and our decisions and statements of reasons on content — and the copy we are mailed of a message sent through the contact form, a bug report or feedback, or a supplier's claim on a maker's page. |
| Anthropic — the AI assistant and the MCP connector | Only when you use them: the messages you send, and what the assistant reads to answer them — your open design, your saved buildings, the list of your weather years, the catalog and the published offerings, and the results of the solves and runs it starts. If you store your own AI provider key, the traffic is billed to your account instead of ours. |
| Hetzner — hosting (Germany) | What we collect (above), at rest in our database. The application and its database run on servers in the EU. |
jsDelivr — a public file network (cdn.jsdelivr.net) | Nothing, unless a name shown in the 3D view is written in a script the app's own label font does not carry. That font is served by the app itself and holds Latin, Greek and Cyrillic; for a label with other characters — Chinese, Japanese, Korean, Arabic and the like — your browser fetches the missing glyphs from jsDelivr directly. The request names the Unicode ranges it needs, not the text, and reaches them from your address, as any download does. |
Outside the EU. What reaches Clerk, Stripe, Resend, Google or Anthropic may be held or read in the United States, where each of them, or the group it belongs to, is based.
Outbound weather data comes from PVGIS (European Commission, JRC). No personal data is sent to obtain it.
How long we keep it
Product notes are opt-in: nothing is mailed to an account that has not ticked the box. Ticking it stores your address and an unsubscribe token; every note carries a one-click unsubscribe link that deletes both, and the box on your account does the same. Notes are written and sent by hand when there is news — there is no schedule, no open or click tracking, and the sender sees a count of subscribers, not the list.
Saved systems, buildings and supplier listings are kept until you delete them or close your account. So are the weather years you import or generate: each belongs to the account (or the workspace) that made it, is listed for nobody else, and stays until you delete it — an account holds up to 25 of its own. A typical year fetched for a city comes from PVGIS, is shared by every account and holds nothing of yours. Simulation runs are pruned automatically to your plan's limit, oldest first. A run in progress is filed as one — whose it is, what it was charged and which start of the server filed it — so that a run the server loses to a crash is ended and its tokens returned when the server starts again; the entry ends with the run. A building run you share by link is readable by anyone holding that link — the results, under the names you gave the building, its rooms, its machines and their systems, and the name of the weather year it ran on; never the design — until you stop sharing it or the run is deleted; the link is a random token, not the run's id, and its views are counted like any page's, so the page counts above keep that address for their 400 days, where it opens nothing once sharing has stopped. Quote requests are kept per supplier up to a fixed retention cap (currently 500), after which the oldest are deleted. Bug reports and feedback filed in the app — including the design snapshot the form attaches when you leave that box ticked — are kept in our database up to a fixed cap (currently 500 across all users), oldest deleted first; the copy mailed to us, which carries no snapshot, is kept for as long as we need it to deal with the report. Messages sent through the contact form are kept up to a cap of their own (currently 500), oldest deleted first; the copy mailed to us is kept for as long as we need it to answer you, and to keep a record of a report of illegal content and what we did about it. Email sent to leads@, bugs@ or privacy@hvacbuilder.app, our answers to it, and the copy of a supplier's claim on a maker's page mailed to us are kept in those mailboxes for as long as we need them to answer you, to do what was asked and to keep a record of it, and are then deleted. OAuth access tokens expire after 8 hours and refresh tokens after 90 days; revoking a connection deletes them immediately. A connector's registration that nobody goes on to authorise is deleted after 24 hours. The id of each payment event our payment processor sends us is kept for 30 days — so that an event delivered twice is applied once — and holds nothing else. An API key exists until you revoke it. Deleting an item in the app deletes the record, not a flag on it. A nightly backup of the database, kept on our server only to recover the service from a failure, holds a copy of what was there that night; each backup is deleted by the first nightly backup to complete once it is more than 14 full days old.
Your choices
In the app you can delete any saved system, building or run, revoke any API key, and disconnect any connected app — each takes effect immediately. A weather year of your own is deleted the same way through the API or an MCP connector (DELETE /api/weather/{id}, delete_weather_dataset). To close your account and have everything associated with it deleted, or to request a copy of it, email privacy@hvacbuilder.app and we will action it. If you are in the EU or UK, the GDPR rights of access, rectification, erasure, restriction, portability and objection apply, and you may complain to your local supervisory authority.
Security
Everything is served over HTTPS. API keys and OAuth tokens are stored only as SHA-256 hashes — a database copy would not yield a usable credential. Administrative actions require an interactive signed-in session, never a long-lived key. Access to another account's data is refused by owner scoping, not by hiding links.
Children
HVAC Builder is a professional engineering tool. It is not directed at children and we do not knowingly collect data from them.
Changes
If this policy changes materially we will update the date above and, for changes that affect how your data is used, tell account holders by email. Questions: privacy@hvacbuilder.app.