Skip to content

Security

How we protect your data

You bring the data your business runs on. Here is exactly what we do with it, and what we never do. Each practice on this page has an automated check; the result below comes from the live service.

14 of 15 checks passed on the live service (Sep 30, 2026).

Encryption

Encrypted, with a key per account

  • Encrypted with a key of your own

    Your tables, photos, sounds, trained models and batch-run outputs are encrypted with AES-256-GCM, with a key unique to your account, before they reach our database or storage. That key is itself sealed by a master key that we can rotate without downtime. Files on AWS are encrypted a second time by AWS KMS.

  • Your credentials, sealed

    The passwords and keys you give us to reach your database or apps are encrypted as soon as they arrive, never shown again, never written to logs, and only decrypted in memory while a connection is open.

  • Delete means gone

    Deleting your account deletes its data and destroys your account's key, so any copy left in a backup can no longer be read by anyone. Backups themselves are kept only for a limited time, and you get an emailed receipt of what was deleted.

Connections

We read only what a model needs

  • Read-only access, nothing more

    Connections to your apps (Shopify, Stripe, QuickBooks, Salesforce…) only read, and only the fields a model needs: ids, dates, amounts, statuses and places. Your customers' names, emails and phone numbers are never requested. For databases we ask for a read-only user and give you the exact grant to create, and we only connect to addresses on the public internet, never to private networks.

AI and personal data

People stay out of the AI and out of public view

  • People stay out of the AI

    Columns that hold people (names, emails, phone numbers) never show their values to the AI assistant, and emails and phone numbers are masked wherever else they appear. Personal columns are also left out of models: they never help a prediction.

  • Nothing personal gets published

    Publishing a table to the community is refused while any column looks like personal data (names, emails, phone numbers, addresses, card or government numbers), checked on the column names and on the values themselves.

How the AI assistant uses your data: to pick the right column and explain results, it sees column names, types and a few example values of non-personal columns, sent to the AI provider listed below through its API. With your own AI key, calls go to your provider under your own agreement with it. Models you train are trained on our servers, not by the AI provider.

Sign-in

Your account, well guarded

  • Passwords done right

    Passwords are stored as Argon2id hashes, the method OWASP recommends first; we never see them. A new password is refused when it is one of the most used ones or contains your email name, checked on our server: your password never leaves it.

  • Two-step sign-in

    Turn on two-step sign-in with any authenticator app. Each code works once; recovery codes cover a lost phone; a device you trust skips the step for 30 days. Even a password reset by email still asks for the code.

  • See and end every session

    Your account page lists every browser and device signed in, with when and from which network. Sign one out, or everywhere else at once: it takes effect immediately. A sign-in from a new device emails you.

  • Guessing is slowed down

    After a few wrong passwords, each new try waits a little longer, per account and per network. The real owner is never locked out: resetting the password by email lifts any pause.

Your control

You see everything, you decide what stays

  • You decide how long

    Choose how long prediction logs, assistant conversations and older table versions are kept, or keep them. Nothing is deleted by surprise: the first clean-up is announced a week ahead. Download everything your account holds, in one ZIP, at any time.

  • An activity log you can trust

    Sign-ins, API keys, connected sources, exports, downloads and models put live are written to your activity log with the device and network. Each entry is chained to the one before by a hash: an entry changed or removed shows at once.

  • Keys that do only their job

    API keys can be limited to reading data, making predictions or acting on the account, given an expiry (with a reminder a week before), and restricted to your office or servers' addresses. Rotating a key keeps its limits.

Infrastructure

Built and checked like it matters

  • Hardened by default

    Everything is served over HTTPS with strict security headers (HSTS, no framing of our pages, no content sniffing). Every change to our code is scanned for known vulnerabilities in dependencies, leaked secrets and risky code patterns, and a test tries every route with another account's ids to prove no account can reach another's data.

  • Logs without secrets

    Our server logs never keep secrets or personal data: keys, tokens, passwords in addresses, emails and card-like numbers are replaced before a line is written.

What we keep

What we collect, and why

  • Your account: name, email and a password hash, to sign you in.
  • The tables you upload or connect, the models you train and their results, to do the work you ask for.
  • Usage counts (calls, rows) to apply your plan's limits, and your activity log, for you to read.
  • Payment details are handled by Stripe; we never see or store card numbers.
  • No advertising trackers. We never sell or share your data, and never use one customer's data to serve another.

Who else touches your data

The services we rely on, and what each does.

ServiceWhat for
RenderHosts the web app and its database.
Amazon Web ServicesFile storage and larger training jobs, in Canada (ca-central-1) by default.
StripePayments and invoices.
AI model providers (via OpenRouter), or your ownThe AI assistant's answers (see how it uses your data above).
Amazon SESAccount emails (sign-in alerts, password resets, notices).

What we don't claim

We don't hold a SOC 2 or ISO 27001 certification yet. Instead of a badge, every practice above is checked automatically on the live service. If your company needs a security questionnaire or a data processing agreement, write to us.

Found a vulnerability?

Email security@datazimuts.com with the details. We read every report, and ask you to give us time to fix an issue before making it public. Good-faith research that avoids other people's data and service disruption is welcome.

For anything else: hello@datazimuts.com.

Use your own AI key

Once today's free allowance is used up, AI features can run on your own provider account.

Kept in this browser tab only (cleared when you close it) and sent with each AI request. Our servers use it for that request and never store or log it.