Rate Limiting
The rate limiting guardrail prevents AI agents from sending too many emails too quickly. A malfunctioning agent or infinite loop can send thousands of emails in minutes, destroying your domain reputation and overwhelming your customers. Rate limiting stops this before the damage is done.
| Default | Enabled |
| Analysis method | Deterministic counting at SMTP level |
| Risk levels | Green / Orange / Red (depends on breach action) |
Three-tier limits
Rate limiting operates at three levels simultaneously:
| Tier | Default limit | What it controls |
|---|---|---|
| Global per hour | 100 | Total emails across all recipients |
| Per domain per hour | 50 | Emails to a single domain (excludes public providers like Gmail, Yahoo) |
| Per address per day | 10 | Emails to a single recipient address |
All limits are adjustable in Settings > Guardrails > Rate Limiting.
Breach actions
When a limit is exceeded, you choose what happens:
| Action | Behavior |
|---|---|
| Block | Rejects the email at the SMTP level (452 error) — the agent can retry later |
| Queue | Accepts the email but holds it in the review queue |
| Alert and continue | Logs a warning but sends the email anyway |
Warmup mode
New sending domains need to build reputation gradually. Warmup mode starts with tighter limits and automatically relaxes them over time.
| Setting | Default | Description |
|---|---|---|
| Warmup enabled | Yes | Use tighter limits for new organizations |
| Warmup duration | 7 days | How long warmup lasts |
| Warmup hourly limit | 20 | Sends per hour during warmup |
After the warmup period, limits step up to your configured values automatically.
Use cases
Development and testing — Your team is building and testing AI agents. Enable rate limiting with low thresholds to catch bugs and infinite loops before they cause problems.
Production protection — Your agents are live and sending customer emails. Set reasonable limits based on your expected volume and use the block action to hard-stop any runaway behavior.
Gradual rollout — You're deploying a new AI agent to production. Use warmup mode to ramp up sending volume gradually and protect your domain reputation.