Legal documents

Information Security Policy

Last updated: October 8, 2026

Courtesy translation. The Portuguese version is the legally binding one.

This document describes the security controls that UDISOFTBOX CONSULTORIA E TECNOLOGIA LTDA ("INGESTIA") maintains on the ingestia.io platform. It exists to be read by whoever evaluates a vendor — security, legal, procurement — and by any customer who wants to know how their data is protected.

Every measure below corresponds to an implemented, verifiable control. The infrastructure these controls run on, and its certifications, are in section 13.

1. Shared responsibility

INGESTIA protects the Platform. The Customer protects what sits on their side.

INGESTIA's responsibilityCustomer's responsibility
Isolation between customer environmentsWho gets access to the environment, and with which role
Encryption of the credentials the Customer registersPassword strength and second-factor activation by their users
Audit trail of actions on the PlatformReviewing that trail and acting on what it shows
Availability and recovery of the PlatformSecurity of their own data sources, networks and devices
Cost and query guardrailsDeciding what data enters the Platform, and whether it may

A Platform control does not substitute for security at the source. An exposed source stays exposed after it is connected.

2. Access, identity and roles

  • Passwords are stored as a cryptographic derivation (scrypt), never in readable form. INGESTIA has no way to read anyone's password.
  • Second factor (MFA) via authenticator app (TOTP) is available for any account. Once enabled, the code is required on password sign-in as well — enabling it is the user's or the environment administrator's decision.
  • Single sign-on (SSO) through the udiApps account, for those already in the ecosystem.
  • Automated provisioning (SCIM 2.0) for customers who manage users from a corporate directory: create, update and deactivate a user without going through our desk.
  • Roles and permissions: five roles (owner, administrator, member, viewer and organisation administrator) over a closed list of 32 actions. A viewer does not run, export or spend; the check runs on the server, on every action, not in the screen.

3. Isolation between customers

This is the control that matters most in a data product, and so it is the most explicit one:

  • each customer has their own datasets in the analytical warehouse, named from an exclusive prefix;
  • infrastructure permissions are scoped to that prefix;
  • a query that tries to reach an address outside the authorised prefix is refused before it runs — it is not filtered afterwards;
  • the result cache is segregated per customer: one environment's answer is never served to another;
  • the destination of each query does not come from the browser's request, but from the dashboard or model that originated it — a request does not choose which environment to read from.

4. Encryption

  • In transit: TLS across all access to the Platform and to our programming interfaces.
  • At rest, for credentials: everything the Customer registers as a secret — source database password, service key, API token and the Customer's own AI key — is encrypted with AES-256-GCM before it reaches the database. The encryption key lives in a server environment variable and is not reachable from the browser.
  • A secret does not come back. The interface never redisplays a stored secret: at most it shows the last four characters for recognition. Secrets are also never written to execution logs.
  • At rest, for business data: encryption managed by the storage infrastructure, as set out in the Subprocessor List.

5. Audit trail

Every relevant action on the Platform is recorded with author, action, resource, moment and — where applicable — the SQL executed and the cost of that execution. The trail is visible to the Customer inside their own environment and is kept for 730 days, as set out in the Data Retention Policy.

The presence of SQL in the trail has a privacy consequence, and it is declared: a query may contain a filter value that is personal data. That is described in the Privacy Policy, and it is why the trail has a deadline and automatic purging.

6. INGESTIA's administrative access

Saying "restricted access" without saying what it reaches informs nobody. So:

  • the INGESTIA team has an administrative access mechanism that allows opening a customer's environment to provide support and diagnose a problem;
  • it is limited to super-administrator accounts, controlled by a server environment variable — it is not granted through a screen or an invitation;
  • its use is recorded in the audit trail, at the moment the access begins;
  • the Customer is notified automatically inside the environment itself: for seven days after the access, a banner at the top of the screen states that the team came in, when, and how many times. The notice depends on no e-mail and on no configuration by the Customer;
  • it is not used for any purpose other than support, operation and security of the Platform.

The notice is given on the screen of the environment itself, where delivery is certain, and is kept in the audit trail.

7. Resilience and recovery

  • Analytical warehouse: recovery of a previous state for seven days (time travel), which covers deletion or mistaken transformation performed by the Customer themselves. The Customer recovers it in the Data console, by querying the previous state, or support does it on request.
  • File landing zone: object versioning — overwriting does not destroy the previous version.
  • Metadata database: backup and point-in-time recovery, through the managed service described in the Subprocessor List.
  • Health check: the Platform exposes a deep check that reports the real state of each capability. In production, a path with no configured credential fails visibly instead of simulating success — an architectural rule, with a test that enforces it.

8. Execution guardrails

  • Per-query cost ceiling: execution is refused by the engine when it would exceed the maximum authorised data volume, before generating any cost.
  • Named parameters in SQL generation — a defence against injection through a filter value.
  • Rate limiting on sensitive points (authentication, public forms, integration interfaces), distributed across instances.
  • Browser security headers, including a content security policy (CSP) per application area and framing restrictions.

9. Development and change

  • All Platform code is covered by an automated suite run before every release, with specific guards for the rules that have already caused damage — among them the byte-for-byte freezing of the query engine's output and the prohibition on any path that fakes success in production.
  • Database schema change is additive and versioned in a migration.
  • Formats saved by the Customer (dashboard, filter, theme) use a versioned envelope with mandatory reading of the previous version: an update of ours does not destroy their configuration.

10. Vendors

The role, purpose, data involved and country of each vendor are set out in the Subprocessor List, kept public and up to date. A new material vendor is announced at least 15 days in advance, as set out in the Data Processing Agreement (DPA).

When the Customer registers their own AI key, the AI provider it serves is not an INGESTIA subprocessor: the contract and the prompt retention policy are between the Customer and that provider.

11. Security incidents

Once an incident involving the Customer's personal data is confirmed, INGESTIA notifies without undue delay and, whenever feasible, within 24 hours, with whatever is available: nature, estimated categories and volume, affected data subjects, containment, risks and point of contact — and updates as the investigation advances.

Communication to the Brazilian data protection authority (ANPD) and to data subjects is the Customer's decision, as controller, within the applicable regulatory deadline. INGESTIA supports the investigation and the communication.

12. Penetration testing and flaw reporting

Penetration testing, scanning or load testing against the Platform requires prior written authorisation — requests to seguranca@ingestia.io.

Anyone who finds a flaw in good faith should report it through the same channel. INGESTIA does not treat a good-faith report as a breach of the Acceptable Use Policy. The procedure and the commitment are in the Vulnerability Disclosure Policy.

13. Infrastructure and certifications

The Platform runs on Google Cloud infrastructure, São Paulo region (BigQuery, Cloud Storage and Cloud Run for the data lake and processing), with the application and the metadata database also in Brazil. The full list of suppliers and regions is in the Subprocessor List.

  • Physical, network and infrastructure controls are Google Cloud's, audited by third parties and certified under ISO/IEC 27001, 27017 and 27018 and the SOC 1, SOC 2 and SOC 3 reports. The reports are obtained directly from the provider, on the Google Cloud compliance portal.
  • Platform controls (identity, roles, isolation between customers, credential encryption, auditing, query and AI guards) are those of sections 2 to 12, operated by INGESTIA. The certifications above belong to the infrastructure provider; INGESTIA is responsible for the Platform controls described in this document.
  • Own cloud: in the model where the data lake runs in the customer's Google Cloud project, the data stays under the customer's account and policies, including the cloud's own keys, auditing and retention.
  • Credential encryption key: kept as a protected secret of the runtime environment, separate from the database holding the encrypted credentials and not reachable from the browser. Rotation follows an operational procedure that re-encrypts every credential.
  • Penetration testing by the customer: authorised upon prior request, under section 12.

14. Contact

Security: seguranca@ingestia.io · Privacy and data subjects: privacidade@ingestia.io · Data Protection Officer: Encarregado de Dados (DPO) — encarregado@ingestia.io