TTS · Reliability

Does it work outside us-east-1?

Production-grade availability and latency across regions. Tail latency, cold-start penalties, burst recovery, and rolling 30-day uptime — the metrics voice-agent buyers actually care about.

v2 — needs probe fleet

Reliability data only becomes meaningful with continuous, multi-region probing — eight cloud functions polling every minute for 30+ days. That probe fleet is dedicated infrastructure that lands in v2.

  • Tail latency by region

    p50 / p95 / p99 from 8 regions: us-east, us-west, eu-west, eu-central, ap-south, ap-southeast, sa-east, af-south. The "does it work in Mumbai?" question, answered with data.

  • Cold vs warm

    First-request penalty after instances scale to zero. Reveals published TTFB numbers that only hold under sustained traffic.

  • Burst behavior

    Spike to 10× normal QPS for 30 seconds. What fails first — and how long until recovery. Reproducible failure-mode receipts.

  • 30-day uptime

    Continuous polling against a target SLA. Production-grade availability with statistical power.