Service-Level Agreement.

99.9% availability per calendar month. That is no more than 43 minutes of unplanned downtime per 30-day month, distributed however it falls.

Measured from an independent external probe that hits /api/health every 60 seconds from at least three geographic regions. A minute counts as down only if two or more regions agree.

When monthly availability falls below 99.9%, the customer is entitled to a service credit applied to the next invoice.

99.0% – 99.9%

10% of monthly fee

95.0% – 98.9%

25% of monthly fee

Below 95.0%

50% of monthly fee

Maximum total credit: 50% of the monthly fee for that calendar month. Credits are the sole and exclusive remedy for availability failures, except in cases of willful misconduct.

Target end-to-end turn latency below 500 ms, measured at the 95th percentile across the customer’s voice calls in a calendar month.

p95 is the right statistic for real-time voice: occasional outliers are tolerable because the conversation absorbs them; chronic > 500 ms degrades the call feel.

Customer admins see real measurements at /admin/latency. When p95 exceeds 500 ms in a month, the customer receives a written acknowledgement, root cause, and mitigation plan within 15 business days.

Sundays, 09:00 – 11:00 UTC (02:00 – 04:00 US Pacific, Saturday night). Announced at least 7 days in advance via the customer admin email and the status page. Scheduled maintenance does not count against availability.

Postmortems are published to the customer admin within 5 business days of Sev 1 or Sev 2 resolution. Public postmortems land at /retros.

• Failures caused by the customer's own systems, configuration, or third-party integrations.
• DDoS or other malicious attacks targeting the customer's tenant specifically.
• Scheduled maintenance.
• Force majeure (natural disasters, war, sustained internet backbone outage).
• Sub-processor failures we notify the customer of within the response windows above and work in good faith to remediate.