Security & trust
Inbox handles your customers' messages and your store's order data. Here is exactly how we keep both safe — and why a human is always the final check.
The short version
Safe with your store and your customers
Six controls do most of the work. The approval gate is the one that matters most.
Approval plus preflight
Sensitive sends, refunds, replacements, discounts, and policy exceptions wait for a human decision after grounding and recipient checks run.
Protected access
TOTP two-factor login is available, OAuth tokens are encrypted at rest, and secrets are never exposed to the browser.
Tenant isolation
Every brand's data is isolated at the database level with row-level security so queries cannot cross stores.
Untrusted customer input
Customer email, attachments, and order notes are treated as data, never instructions to the AI, to reduce prompt-injection risk.
AI health and evals
Model calls, success rates, grounded send proof, scenario gates, and repairable evidence gaps are visible in the analytics workspace.
Searchable audit proof
Approvals, edits, sends, actor IDs, grounding proof, and repair actions are reviewable from the audit log.
The detail
Controls, mapped to the real product
These are the controls represented in the product and delivery pipeline. Certification roadmap items are separated from live controls.
Human approval gate
Nothing reaches a customer until a human taps approve. Every draft is also re-validated whenever it is edited, so a hand-tweaked reply can never bypass the safety checks. A live audit counter proves zero replies were sent without approval.
Tenant isolation
Every brand's data is isolated at the database level with Postgres row-level security on every tenant table. A query can never return another store's data — RLS is the backstop beneath application-level scoping.
Secrets & tokens
OAuth tokens (Shopify, Gmail, Zoho) are encrypted at rest and never logged. No secrets live in code — they are injected as environment variables, and the service fails fast if its encryption key is missing.
Untrusted input
Customer email is treated as data, never as instructions. Every draft is scanned for PII, policy violations, and prompt-injection markers before it ever reaches your queue.
Webhooks & integrity
Inbound webhooks are signature-verified and de-duplicated by event id before processing — replays and forged events are rejected. State changes are CSRF-protected and same-origin checked.
Hardened by default
Strict security headers (CSP, HSTS, X-Frame-Options: DENY), TOTP two-factor login, Cloudflare Turnstile + rate limiting on auth, and automated dependency-vulnerability and secret scanning in CI on every change.
Found a vulnerability? See our vulnerability disclosure policy. Have a security question, or need this for a vendor review? Email us — we respond to disclosure reports and security questionnaires directly.
A trust posture your buyer can actually inspect
Inbox sells an AI employee, so trust cannot be a footer badge. The homepage now shows the exact controls behind approval-first automation, data boundaries, OAuth safety, auditability, and the enterprise roadmap.
Bring every support decision into one workspace.
Inbox drafts with Shopify context, knowledge proof, preflight checks, and approval routing before anything reaches a customer.