# Security

_Last updated: 21 July 2026_

Security is core to how **alia.help** (operated by **Lumus.gg LTDA**, CNPJ
40.325.568/0001-05) is built, because the Service connects to production
databases. This page describes the measures actually in place. It is a summary,
not a warranty; see our [Terms of Service](/terms).

## Read-only by construction

Every query the Service runs against a connected database is contained by five
independent layers:

1. The connection sets the session to **read-only** and applies a statement
   timeout on connect.
2. Every query runs inside a transaction explicitly marked **read only**.
3. Results are streamed with a hard **row cap**.
4. A result **byte cap** (~2 MB) bounds how much data any single query returns.
5. Only **single statements** are accepted; multi-statement SQL is rejected by
   the database protocol.

The result: your agent can explore your data, but it **cannot write to, alter,
or delete** anything in your database. Write and schema-changing operations are
deliberately outside the product's scope.

## Credentials and secrets

- Database passwords are **encrypted at rest** using AES (via Cloak) before
  being stored.
- API and MCP access tokens are stored only as irreversible **SHA-256 hashes**.
- Sign-in is **passwordless** (magic link), so no user passwords are stored.

## What is sent for AI processing

Only **schema metadata** (table and column names, data types, and constraints)
is sent to our AI sub-processor to build your semantic catalog. Your table rows
and cell values are **not** sent for this. See our
[Sub-processors page](/sub-processors).

## Network and hosting

- The Service is served over **TLS**.
- Application servers and databases are hosted with **Hetzner** in the EU
  (Germany).
- The Service connects to your database from a fixed outbound IP address —
  **5.78.219.67** — which you can add to your database firewall's allowlist.

## Tenant isolation

Each user's hosted CSV workspace is an isolated schema, and schema introspection
is scoped to that user's own schema so table metadata does not leak across
tenants.

## Responsible disclosure

If you believe you have found a security vulnerability, please email
**privacy@alia.help** with details. We appreciate responsible disclosure and
will investigate promptly.