What we protect, how AI is governed, and what we won't claim.
This page is written for the person filling in the questionnaire. It describes the controls that exist today. Where something is a roadmap item, it says so.
Credentials sealed
Mailbox and AI keys are encrypted at rest with AES-256 and scoped to a single workspace. Reset, verification and invite tokens are hashed; a password change revokes other sessions.
Tenant isolation
Every table carries the workspace; isolation is tested in an automated suite on each deploy. Public endpoints are rate-limited and request logs are redacted.
Audit trail
Every AI proposal, approval, send decision and guardrail verdict is written to an append-only ledger with who, what and why.
Passwordless sign-in
Email OTP sign-in by default; no password policy for you to run. Second-owner and hardware-key options for owner accounts.
Outbound safety
DNS-resolved SSRF guard on webhooks and mailbox hosts, sandboxed rendering of user-written HTML, CSV formula-injection protection.
Backups
Nightly encrypted database backups with an age check that alerts if a backup is missed; off-box copies.
AI proposes. A person approves. Everything is logged.
Delegated, not direct
Model output is a draft. A person — or a rule you wrote — approves it against scope and policy. Only then does deterministic code write it, with an audit line.
Data boundaries
Agents see only the workspace records the requesting user can see. Nothing is used for training. Keys belong to the customer and can be removed at any time.
Independent verification
After execution, a verifier compares what happened with the approved plan and reports drift. Owner-side QA agents run nightly against the platform.
Sending safety
- Sequence Guardrail with a Sequence Risk Score on every send
- Dispatcher enforces the lower of your cap and the guardrail's cap
- One-click unsubscribe headers and shared suppression lists
- Recipient-local windows and regional rules
- India: AI outbound calls only from TRAI-compliant numbers via licensed carriers
Operations
- Single dispatcher with advisory locks — no double sends
- Secret rotation with a grace period
- Dependency CVEs patched on a regular cadence
- security.txt published for responsible disclosure
- Written answers to security questionnaires
Questions, answered
Are you SOC 2 or ISO 27001 certified?
No, and we say so plainly. We answer security questionnaires in writing, describe the controls below accurately, and will not claim a certification we have not earned.
Where is data hosted?
On dedicated infrastructure operated by XDQ Labs, with nightly encrypted backups and off-box copies. Enterprise and Scale plans can be deployed on dedicated or private infrastructure by agreement.
Do you train AI models on customer data?
No. AI features run on the customer's own provider key, and nothing from a workspace is used to train any model, ours or anyone else's.
Can I delete my data?
Yes. Contacts, deals and campaign history export at any time, and a workspace purge erases records — including message evidence and decision ledgers — on request.
Do you sign a DPA?
Yes, on Enterprise and above, alongside a security review. Smaller plans are covered by the standard terms and privacy policy.
Need it in writing?
Send the questionnaire. We answer every question, including the ones where the answer is “not yet”.