Written to be checked, not admired.

Tenant isolation is enforced by the database, not by our code

Every table that holds your data carries row-level security. Every request runs as a non-owner database role inside a transaction that sets your tenant id; the policies refuse any row that is not yours.

The isolation is mutation-tested: our test suite disables the policies and asserts that eleven isolation tests go red. A privileged role is refused by the tests themselves, so a passing suite cannot be a false comfort.

Credentials and keys are encrypted at rest

Mailbox passwords, API keys for your email providers, your AI key if you bring your own, and MFA secrets are encrypted with a server-side key before they are stored. They are never returned to a browser; the settings screens show only that a secret is set.

Sign-in supports passwords with TOTP two-factor authentication, single sign-on via your identity provider, and passwordless links. Sessions are HttpOnly cookies over HTTPS.

AI reads your data only to do the job you asked

AI drafts, scores and triages on demand; a person approves before anything is sent or submitted. Every AI feature degrades to a named state when its key is absent rather than failing silently.

You can run all AI on your own Anthropic key. When you do, no prompt from your tenant is sent under ours. Prompts are not used to train models by either party under the API terms we operate on.

Reviewed, and the findings closed

A pre-launch security review in September 2026 covered server-side request forgery, authentication spraying, host-header takeover, extension token handling, AI spend limits and response headers. Every finding was fixed and the fixes are in the audit trail.

An immutable audit log records who changed what, in every tenant, for org admins to read.

Where your data lives, and who else touches it

Application and database run on a single-tenant-controlled Azure virtual machine in the United States. Backups stay in the same region.

Sub-processors: Microsoft Azure (hosting), Anthropic (AI, unless you bring your own key), Microsoft 365 or SendGrid (transactional mail you do not send from your own mailbox), Stripe (billing), Sentry (error monitoring; no message bodies). We do not sell or share your data with data brokers, and we never send tenant data to a contact-data vendor.

Data retention is configurable per tenant in Settings → Data retention. On request we export or delete a tenant's data in full.

Outreach protects your sending reputation

You send from your own domain through your own mailbox or provider. Volume is paced by a per-tenant daily cap and a warm-up ramp; every email carries the legally required footer and unsubscribe; suppression lists are honoured; bounces stop further sends.

What we do not have yet, and how to reach us

We do not hold a SOC 2 report or a FedRAMP authorisation. The controls above are real; the third-party attestation is a purchase we will make when a customer's procurement needs it. If yours does, tell us and we will schedule it.

Found a vulnerability? Write to hello@velocapture.com with the word security in the subject. We acknowledge within one business day and do not pursue good-faith researchers.