🛡
Nothing posts without your approval (default)
Every AI draft — review reply, social post, email, customer SMS — lands in your manual approval queue first. The AI cannot speak for your business until you say yes. Auto-send stays off unless you deliberately enable it for one specific content type, and even then only after you have approved enough of them by hand.
- Approval queue gates every generated piece of content
- Edit-in-place before approving — your changes train the AI
- Bulk approve / reject is opt-in per content type, never default
🧠
Auto-approve is opt-in AND earned
You can opt specific content types into auto-approve to skip the queue — but only after you've manually approved 10 of that type. The AI has to earn trust on your real voice before going autonomous.
- 10-approval threshold per content type (configurable up)
- Hard blocklist — crisis responses, price changes, hiring posts NEVER auto-approve regardless
- Reversible any time in Settings → Auto-approve
🔒
AES-256 encryption for every stored credential
API keys, OAuth tokens, app passwords — everything sensitive gets encrypted at the application layer before it touches the database. Plus full-disk encryption on the underlying Postgres host.
📋
Append-only audit log
Every security-sensitive action (invites, integration changes, account exports, plan changes) writes an immutable row with actor + IP + user-agent. Available to admin via /admin/audit-log.
👤
You own your data, period
GDPR Art. 20 (data portability) and Art. 17 (right to erasure) are first-class features, not legalese.
- One-click JSON export at /account/export
- Account deletion at Settings → Danger Zone cascades through every table
- Audit log records preserved with actor ID nulled (security retention)
🛟
AI Receptionist warmup gate
When the AI auto-replies to customer SMS, the first 50 messages stay in your manual queue even if you opt into auto-send. You verify tone + facts on real customers before the AI goes live.
🚫
Business-only AI
The Anthropic Claude prompts are constrained to business operational tasks. The AI won't engage with off-topic requests, won't hallucinate prices it doesn't know, won't share customer data with anyone outside your account.
🌐
No silent data sharing
We don't sell or share your generation data with third parties. Your review content + customer messages go to Anthropic per-request (as part of the AI generation) and are not used to train their models.