How we protect your data

Deny-by-default access control

Every table enforces Row-Level Security with no implicit access. Requests are authenticated and scoped to the caller's own account, so a request scoped to one account is not able to read another's. These policies are covered by ongoing security review.

Signed webhooks & billing integrity

Inbound webhooks (email, payments) are signature-verified and fail closed. Billing events are idempotent, so a payment or credit grant is applied exactly once.

SSRF & egress protection

Any server-side fetch of a user-supplied URL is validated against private, loopback, link-local, and cloud-metadata ranges, with redirects re-checked at every hop.

Hardened HTTP surface

Strict Content-Security-Policy, HSTS with preload, clickjacking protection, and a locked-down permissions policy on every response.

Least-privilege secrets

Service-role database keys and third-party API keys live only in server environments. Client writes to sensitive fields (credits, billing) are rejected.

Ongoing security reviews

We run internal security reviews across authentication, access control, injection, and dependency hygiene, and remediate findings on a tracked cadence.

Where the data lives

The warehouse, BI tenant, and any integration infrastructure are provisioned in your cloud accounts under your ownership. Your operational data does not flow into the Valentino AI self-serve product.

Who owns the credentials

You own the Snowflake, DOMO, telephony, and TMS credentials. Access granted to us or to an implementation partner is named, least-privilege, and revocable by you at any time.

TMS and operational data

The assessment runs on interviews, documentation, sample reports, and screen-shares — not a data dump. Where a question genuinely needs read access, we ask for it explicitly and scope it narrowly.

Environment separation

Build and test work happens against non-production data or a separate environment wherever the source systems support it. Where they don't, that constraint is written into the engagement plan rather than worked around silently.

Retention and deletion

Retention windows, deletion triggers, and what happens to working copies at handoff are agreed in writing before access is granted, and recorded in the engagement's security schedule.

Incident responsibility

Who detects, who notifies, who remediates, and in what timeframe is defined per engagement and scoped to the access actually granted — not asserted generically here.

Model providers

LLM calls route through OpenRouter to the model configured for that workflow. Which provider handles your data, and what is redacted before it leaves your environment, is part of the architecture we agree up front.

Voice calls

Where a voice agent is in scope, call disclosure, whether calls are recorded, where recordings and transcripts are stored, and who can access them are decided with you before the agent answers a single call.