Seal

Legal

Privacy policy

What app.aqta.ai and Seal collect, why, who processes it for us, and how long it is kept.

Last updated 18 October 2026

In short

  • For AI traffic sent through Seal, your organisation decides how the data is used; we act on its instructions.
  • Seal does not store your requests or the model's responses, only a signed record with a one-way fingerprint, and metadata.
  • Signed records are hashed into a public log that publishes hashes only. We do not delete signed records today.
  • Seal and our database run on Google Cloud in Belgium; some processors handle data in the United States.
  • Cookies only to sign you in and keep your account secure. We do not sell personal data.
  • Questions or requests: hello@aqta.ai.

A plain restatement; the full text below is what applies.

1. Who we are

Aqta Technologies Limited (CRO 807530), 20 Harcourt Street, Dublin 2, D02 H364, Ireland, runs app.aqta.ai and Seal. We decide how account, billing and support data is used. For AI traffic your organisation sends through Seal, your organisation decides, and we process it on its instructions under a data processing agreement, available on request.

2. What we collect

  • Account: your email address, your name if you give it, and your organisation; if you sign in with Auth0, Google or GitHub, that account's identifier. Email-code sign-in stores a one-time code for 10 minutes.
  • Sign-in sessions: when you signed in, your IP address and browser, so you can see your active sessions and sign out of them.
  • Payments: Revolut handles your card, and receives your email address, the amount and an order reference.
  • Use of Seal: signed records and operational metadata (section 3).
  • Human-review requests: your organisation name and email address, sent to our team as an internal Slack alert.
  • Review packs: a reviewer who downloads a pack from a docket gives a work email address, which we record with the time and browser; the pack's owner can see how many downloads there were.
  • Technical: visit and page-speed counts (Vercel Web Analytics and Speed Insights, which set no cookies) and server logs.

3. What Seal stores

Seal inspects the content of your AI requests in transit to apply your policies (for example PHI detection, loop detection and spend limits), then discards it. We do not store request content or the model's responses. For each request Seal keeps:

  • A signed record (ATTESTATION-v1, signed with Ed25519): the time, the policy applied, the outcome (allowed or blocked) and a one-way SHA-256 fingerprint of the request. An agent-action record (ACTION-v1) holds the time, your organisation's identifier, the session identifier, the agent and tool names your system sent, fingerprints of the arguments and the declared intent, the policies applied and the outcome.
  • Operational metadata: timestamps, model name, token counts, cost, the policy outcome and reason, and the IP address and user agent of the system that called Seal.
  • Declared intent: for agent sessions, the intent your agent declares when the session opens, in plain text, with the session.
  • Threat patterns: when Seal detects a threat, the request's fingerprint and Seal's detection details, never the request text.

A fingerprint cannot be read back, but it can confirm a guessed input, so we treat records and patterns as pseudonymous personal data, not anonymous data. Aqta API keys are stored only as one-way hashes. Model-provider keys you add are stored encrypted, because Seal must use them to call the provider for you.

4. The public transparency log

Every signed record and agent-action record is also a leaf in Seal's public transparency log, which publishes hashes and signed tree heads only, never content or account details. A leaf hash covers the whole record and its signature, so it cannot be matched to anything without a copy of the record. The log is append-only and computed from the stored records, so removing a record would break proofs others already hold. Today we do not delete signed records, and published tree heads cannot be withdrawn. How we handle requests to erase or correct a record is under review.

5. Shared threat patterns

Sharing is off unless your organisation switches it on, and we share only on its instruction. When it is on, your deployment contributes threat patterns to, and receives them from, other Seal deployments. You can use Seal without sharing. A shared pattern holds the request's SHA-256 fingerprint and Seal's detection details (threat type, severity, model and confidence), never request text, your organisation name or user identifiers.

6. PHI Guard (Health pack only)

PHI Guard is an optional pack for healthcare deployments. When enabled, it scans requests in transit for Protected Health Information (PHI) categories defined under HIPAA, blocks or redacts before the model call, and records a metadata-only finding (categories detected, action taken, hash of the redaction). Seal does not store PHI text; the signed record for a PHI Guard decision holds the policy outcome and category codes only. Where PHI Guard is deployed under a Business Associate Agreement (BAA), the BAA governs and overrides this policy's defaults. Ask hello@aqta.ai for the BAA template.

7. Reviewer verdicts (review.aqta.ai)

At review.aqta.ai, a prototype labelled as one wherever it appears, a reviewer signs a verdict on a receipt with a credential from their own wallet, disclosing one attribute, today a verified email address, after the wallet asks them to approve it. Aqta does not issue the credential and never sees a password.

We store the verdict, the fixed question, the receipt, the signed binding message and the disclosed attribute, which identifies the reviewer, because an anonymous verdict is worth nothing. Each verdict sits in a docket with an unguessable identifier, and anyone holding that docket link can read it, including the disclosed attribute, without an account. That is deliberate, so a third party can check it, and the page says so. The basis is consent, given at the disclosure step, which you may withdraw. Verdicts are kept while the review flow runs and deleted on request: write to hello@aqta.ai with the docket identifier.

8. Why we use it

  • To provide Seal and the dashboard, analyse your AI request patterns, take payments, manage subscriptions, give support and tell you about service updates (contract; for AI traffic, on your organisation's instructions)
  • To detect and prevent security threats and abuse, and to improve the service (legitimate interest)
  • To keep billing records (legal obligation)

9. Who processes data for us

  • Google Cloud (Belgium): Seal, signing and our database; where we provide model access ourselves, Gemini on Vertex AI, at Google's global endpoint
  • Vercel (functions in Dublin): the dashboard, and visit and page-speed counts
  • Auth0 (EU tenant), and Google and GitHub if you sign in with them: sign-in
  • Resend: email
  • Slack: internal alerts when you request a human review
  • Revolut: payments
  • Amazon Web Services (Ireland): a database from before August 2026 that we can no longer reach

The full list is at aqta.ai/trust. A model provider your organisation reaches through Seal receives the content under your own agreement with it. We also share information when the law requires it, or to protect our rights and prevent fraud. We do not sell, trade or rent personal data.

10. International transfers

Seal and our database run in Belgium, and the dashboard's functions in Dublin. Some processors are based in, or process data in, the United States, including Resend and Slack, and Google runs Gemini at its global endpoint. We are reviewing the safeguards for each of these transfers; ask us for the current position.

11. How long we keep it

  • Account information: until your account is deleted. Sign-in sessions: up to 30 days, or until you sign out.
  • Operational metadata: deleted automatically after the period for your plan (7 days for evaluation access, 30 days on starter, 365 days on pro) or the period in your agreement. Agent sessions started by a model call expire on the same schedule.
  • No expiry set yet, under review: sessions opened by registering an intent, threat patterns, review-pack download records, and signed and agent-action records, which are not deleted automatically (section 4).
  • Billing records: as long as Irish tax law requires; no automatic deletion yet.
  • Server logs: up to 30 days on Google Cloud; Vercel keeps its logs for less.

12. Cookies and browser storage

app.aqta.ai uses only cookies needed to sign you in and keep your account secure:

  • NextAuth session, CSRF and return-address cookies, when you sign in with Auth0, Google or GitHub (up to 7 days)
  • aqta-auth-token, aqta-user-email (your email address) and aqta-user-mode, for email-code sign-in (30 days)
  • aqta-csrf (7 days), and short-lived cookies while you use an invite code (15 minutes) or connect Slack (10 minutes)
  • seal_mission_demo, which remembers access to the mission demo (45 days)

Your browser also stores your dashboard profile (including your email address), display preferences, onboarding progress and, during a payment, the order reference. No advertising cookies. The independent review site at review.aqta.ai sets no cookies.

13. Security

Encryption in transit and at rest, with access controls and authentication, on Google Cloud and Vercel.

14. Your rights

You can ask for access to your personal data, its correction or deletion, a restriction of its use or a portable copy, object to its use, and withdraw consent at any time. Write to hello@aqta.ai; we reply within one month. Where we process data for your organisation through Seal, we pass your request to your organisation and help it respond. You can also complain to the Data Protection Commission (dataprotection.ie).

15. Contact

Data protection: hello@aqta.ai. Security reports: security@aqta.ai. Our address is in section 1. We update this page when what we do changes, and change the date at the top.