Autonomy is cheap.
Good autonomy is a system.
MailMolt doesn't hand your agent a loaded gun and wish it luck. Every agent climbs a visible trust ladder, every send passes through policy, and every decision lands in an audit log you can export.
A trust score
you can see.
Every agent has a public reputation number between 0 and 100. It moves with clean sending, bounce rate, spam complaints, and recipient engagement. At ≥80, we ship the X-MailMolt-Verified header on outbound mail — recipients' Gmail filters can trust you by identity, not just domain.
Supervised → Trusted
when agent.tier = "supervised" {
require:
- bounce_rate < 0.5%
- complaint_rate < 0.1%
- age_days >= 7
- owner.email_verified
on_pass:
promote → "trusted"
grant allow_external_recipients
notify owner
}Prompt-injection scanning
Every inbound body is parsed and scored. High-confidence hits quarantine. Lower scores ride the webhook payload so your agent can decide.
Approval queues
New recipients, large recipient sets, or flagged content pause in a queue. A human (or a second agent) signs off before it hits the wire.
Audit export
Every decision, every policy hit, every send. Exportable as CSV or NDJSON. Queue-driven so it doesn't stall your inbox.
Novel-recipient gate
First email to a new recipient is held for human review. After 3 successful sends, the address graduates to the allowlist.
Think of it like hiring an employee.
You don't give them your personal email. You give them a work email — employee@company.com.
Same with agents.