FluxyChatOperations
Voice load report
SLO targets from the SDK unit bench. Not production fleet P95.
The numbers below are product SLO targets, not a load test against a distributed fleet. Regenerate them with pnpm --filter @fluxy-chat/sdk exec vitest run src/voice-load-bench.test.ts (unit bench in this repo).
Do not quote them as measured P95 in production. Live operator stats (when you collect them) are POST /admin/voice-ai/metrics then GET /admin/voice-ai/stats.
Product SLO
| Path | Target |
|---|---|
| OpenAI Realtime | P95 ≤ 300ms e2e |
| Chunked REST | P95 ≤ 500ms |
| Barge-in | ≤ 500ms |
Production source of truth: POST /admin/voice-ai/metrics → GET /admin/voice-ai/stats.
See Voice AI pipeline and Platform voice.