K99 telemetry & SLAs
What we measure, what we expose, and the SLAs each AI surface targets.
Overview
K99 is a platform, not a black box. This article documents what we measure, what we expose to tenants, and the SLAs each AI surface targets. If a number isn't in this list, it's probably not measured — please tell us what we're missing.
Per-surface SLAs
| Surface | Latency target (p95) | Availability target | Where to check |
|---|---|---|---|
| Storefront render | < 800 ms | 99.95% | status.katura.com |
| Admin page load | < 1.2 s | 99.9% | status page |
| Yen first token | < 1.5 s | 99.5% | /admin/conversations |
| OpenClaw run start | < 30 s | 99% | /admin/openclaw |
| Outbound send schedule | ± 60 s of target | 99.9% | /admin/outbound |
| Stripe webhook ingest | < 5 s | 99.99% | status page |
Metrics exposed to tenants
- AI usage dashboard — tokens consumed by Yen, OpenClaw, and Outbound this billing cycle, with budget caps.
- Conversation outcomes — how many Yen sessions led to a cart, a booking, an escalation, or were abandoned.
- Send health — open / reply / bounce / unsubscribe rates per Outbound campaign.
- Performance budget — Largest Contentful Paint, INP, and TTFB for your storefront, by country, last 28 days.
Incident communication
When K99 declares an incident, three things happen: (1) a banner is shown in affected tenant admin UIs, (2) the public status page is updated, and (3) an email goes to every tenant's declared incident contact. Post-incident, a public postmortem is published within 5 business days.
Tip
Subscribe to the status page at status.katura.com for real-time updates without needing to log in.