Security Tab

Every agent has a Security tab with five independent controls. All settings are saved per-agent and take effect immediately on the next inbound call.

When enabled, credit card numbers, Social Security Numbers, email addresses, and phone numbers are detected and replaced with [REDACTED] before transcript messages are written to the database. The live conversation is unaffected — only the stored text is masked.

Enter E.164 phone numbers (e.g. +12125551234) one at a time. Callers from a blocked number hear a rejection message and the call ends immediately. Numbers are matched exactly — no wildcard or prefix matching.

1. Type the number in the input field. A leading + is added automatically if missing.
2. Press Enter or click Add.
3. The number appears as a removable chip. Click × to remove it.
4. Changes autosave within 3 seconds, or click Save changes immediately.

Block entire countries by ISO 3166-1 alpha-2 code (e.g. US, DE, RU). Twilio reports the origin country in the inbound webhook, so no phone number math is required.

Ten common countries are shown as quick-pick buttons — click to toggle blocked/unblocked. For any other country, type the two-letter code in the custom input and click Add.

Limit how many times the same phone number can call within a rolling time window. Uses Upstash Redis for tracking — if Redis is unavailable, the check is skipped (fail-open) so legitimate calls are never dropped due to an infrastructure issue.

When the limit is exceeded, the caller hears a message and the call ends. The counter resets automatically when the window expires.

Sets the minimum pass rate (0–100%) required when using Run Evals & Publish from the agent builder header. If the agent's test suites score below this threshold, the deploy is blocked and you see a breakdown of failing cases.

Set it to 0 to always publish regardless of eval results. Set it higher (e.g. 90%) for production agents where regressions are costly.