Security & Data Handling – Marapone | Private AI for Construction

The short version

Your documents are uploaded to Marapone to be read — that is how a hosted program works, and we would rather lead with it than bury it. What we control is where they stop: a workspace scoped to your account alone, a model Marapone built and runs, no commercial AI API anywhere in the path, and nothing you upload used to train anything. You can export or delete your material at any time.

No third-party AI vendor in the path

No telemetry, no usage reporting

No cloud dependency post-handover

Marapone has zero post-handover access

No backdoor, no remote access

Delete a project, and its data, on demand

These lines describe everything a plan opens — the programs in the browser and the MaraponeAI mobile app alike. The one thing on this site they do not describe is a data pack, which is a file you download and store yourself, with nothing running and nothing uploaded.

THE FREE TRIAL

A trial is a subscription that has not started billing yet, so nothing about it is a lesser tier of privacy. Documents you put through it are handled exactly as a paying subscriber's are:

Files are transmitted over HTTPS into a workspace only your account reaches, held in isolated storage, and scanned for malware before anything opens them.
They are used only to produce your own output. They are never used to train any model and never shared with any third party.
If you are concerned about sensitivity, you can redact or anonymize a document first. The system extracts meaningful findings from partially redacted documents in most cases.
You can export or delete anything from your account at any time, including after you cancel. Signed access URLs expire after 7 days.
The card is never one of the things we hold. Payment details go straight to Stripe and are stored there, not here — this site never sees a card number, during a trial or after it.

DURING THE BUILD

During the scoping and build phase, you provide sample documents so we can configure the pipeline to your specific document formats. These are handled the same way as any document on a plan — secure transfer, no retention beyond the project, no third-party sharing, and never used to train any model.

All communication about your build — scope documents, specifications, technical details — is kept confidential under the non-disclosure terms in our engagement agreement.

We operate on a strict need-to-know basis during builds. Only the people directly working on your system have access to your documents.

WHILE YOU ARE SUBSCRIBED

The programs run on Marapone infrastructure, so the question is not whether we can reach your documents — we can, that is how they get read — but what constrains it. These are the constraints:

Specifically:

One workspace per account. Your documents and projects are processed in a workspace scoped to your account. No query from another customer can reach them, and no model context is shared between accounts.
No third-party AI vendor. Nothing is forwarded to OpenAI, Anthropic, Google or any other AI provider. The model is ours, on infrastructure we operate.
Nothing trains on your work. Your drawings, quotes and estimates are processed to answer your question and stored so your projects reopen. They are not used to train any model, ours included.
Staff access is exceptional and logged. Our engineers do not browse customer workspaces. Access happens only to investigate a fault you have reported, or where we are legally compelled, and it is recorded.
You can leave with your work. Export any project at any time. If a subscription lapses, projects are held for 90 days so resubscribing restores them, then deleted. Ask for deletion sooner and we action it within 30 days.
Private deployment available. For clients whose contracts require that documents never leave their own infrastructure, we will deploy into your own cloud tenant as an enterprise arrangement. Ask before subscribing — a standard plan cannot meet that requirement.

THE EIGHT PROGRAMS

All eight programs — the five pre-award products (Blueprint Auditor, SpecChecker, the AI Estimator, the Bid Leveler and ScopeGuard) and the post-award suite (FloatGuard, TradeVet and CostLine) — are subscribed to rather than scoped as a build, and every one of them is held to exactly the same standard as anything we build to order. They open in a browser and run on our infrastructure; the mobile app arriving in Q4 2026 is the same service on a smaller screen, on the same plan and under the same handling described here.

Encrypted in transit and at rest, in Canada or the EU. Processing runs on infrastructure we operate in Canada by default, or in the EU for accounts that ask for EU residency. Ask before you subscribe if that matters to your own obligations.
Your project data is yours to take. Estimates, rate sheets, job history, fee-schedule edits and reports all export to PDF, Excel or CSV on demand — no request, no waiting period, no export fee.
Operational logs record that a job ran, not what was in it. Which engine, when, how long, and whether it failed — kept 30 days to fix crashes and size the infrastructure.
No advertising or analytics on your documents. There is no ad SDK, no session-replay tool and no third-party analytics with sight of your project data.
Our own model, end to end. Every document is read by a model Marapone built and runs, on hardware Marapone controls. No commercial AI API is in the path, so no second company holds a copy of your tender package under its own terms of service. Nothing you upload is used to train any model. We will put every line of this in a signed DPA, and name our infrastructure providers, before you commit to anything.

Data packs have a smaller surface still.

A data pack is documents and tables — PDF, CSV and JSON — with no executable, no installer and no macros. There is nothing to run, so there is nothing to phone home, nothing to grant permissions to and nothing to keep patched. Delivery is a link by email; the security question a pack raises is only the ordinary one of where you store the files afterwards, which is entirely on your side.

THE MOBILE APP

The MaraponeAI mobile app, shipping Q4 2026, is the same hosted service reached from a handset — same account, same plan, same engines, same handling. The points below apply to the browser programs equally; they are gathered here because a phone raises two extra questions, about device permissions and about who bills you.

If your requirement is that project documents never leave your own infrastructure, a standard plan cannot meet it — on the phone or in the browser. Deployment into your own cloud tenant is an enterprise arrangement; ask before subscribing rather than after.

For everyone whose requirement is not that, here is what the service does and does not do:

It goes to our server and no further. Your files are not passed to OpenAI, Anthropic, Google or any other AI provider. The model behind the in-app assistant is our own open-weights model on the same infrastructure as the engines, so no third-party AI vendor is ever in the path of your tender documents.
One isolated workspace per account. Each account's documents and projects are processed in their own workspace, and the assistant is only ever given that workspace — there is no query that reaches another contractor's estimate. On a Firm account the seats you add share one workspace deliberately; that is the feature.
Nothing is trained on your work. Your drawings, quotes and estimates are processed to answer your question and stored so your projects reopen. They are not used to train any model, ours included, and not shared with anyone.
Encrypted in transit and at rest, in Canada or the EU. Processing runs on infrastructure we operate in Canada by default, or in the EU for accounts that ask for EU residency. Ask before you subscribe if that matters to your own obligations.
You can leave with your data. Export to PDF, Excel or CSV any time. If a subscription lapses, projects are held for 90 days so resubscribing restores them, then deleted. Ask for deletion sooner and we action it within 30 days.
No advertising or analytics SDKs on the device. The app does not read your advertising identifier, carries no ad or session-replay SDK, and asks for the camera and file access only at the moment you use them. Apple and Google hold your payment details and bill the subscription; we receive the tier, never the card.
Operational logs record that a job ran, not what was in it. Which engine, when, how long, and whether it failed — kept 30 days to fix crashes and size the infrastructure.

The full data-handling detail is in section 06 of the Privacy Policy, and the contractual side is section 13 of the Terms.

SUPPORT ACCESS

When you report a fault, any access to your workspace is scoped to investigating that fault. We ask first, we say what we need to look at and why, and the access is logged. We do not browse customer workspaces for any other reason.

You can decline, and the fault gets diagnosed from logs and your description instead. No standing access is maintained between support requests.

THIS WEBSITE

This website (marapone.com) collects minimal data. We use session cookies required for the site to function, Vercel's anonymous web analytics, and Google Analytics (GA4) to measure aggregate page traffic. Google Analytics does not identify visitors personally and is not used for advertising or behavioural profiling. We do not use advertising cookies, third-party tracking pixels, or behavioural surveillance tools.

Contact form submissions collect only what you provide — name, email, company and your message. This is used solely to process your request. Our demo and forms are protected against bots and abuse by Cloudflare Turnstile, a privacy-focused CAPTCHA alternative that uses no tracking cookies. See our Privacy Policy for full details.

See our Cookie Policy for a complete breakdown of the cookies we set and why.

THE LIVE DEMO

The live demo on the Blueprint Auditor page lets you upload a drawing set and see the engine analyse it in real time. It is the one place on our site where a file you upload is processed on cloud infrastructure rather than locally — so we built it to hold onto as little as possible:

In-memory only. Your file is analysed by a serverless function and discarded the moment the request finishes. It is not stored, not logged, and never sent to any external or third-party AI service.
Strict limits. Uploads are transmitted over HTTPS, capped at 5 MB, and restricted to a file-type allowlist (PDF, CSV, TXT, JSON).
Abuse protection. A Cloudflare Turnstile check plus per-IP and per-browser rate limits guard the demo against automated abuse.
Server-side gating. Locked results are stripped on the server before the response is sent — they are genuinely withheld, not just blurred in your browser.
Use sample files. Because the demo runs in the cloud (unlike a delivered build), please use sample or non-sensitive documents. Highly confidential material should go through a private, locally-handled channel instead.

QUESTIONS OR CONCERNS

If you have specific security requirements — compliance documentation, an NDA before you start, custom data handling agreements, a signed DPA, or deployment into your own cloud tenant — reach out before subscribing. We deal with these regularly.

Contact: general@marapone.com

Ready to start

YOUR DATA.
YOUR WORKSPACE.

Seven free days in a workspace only your account reaches. No commitment, no sales call, and no card details on this site — Stripe holds those.

Start Free 7-Day Trial →

SECURITY &DATA HANDLING