LIVECT
Performance & SLA
Latency, availability, capacity, monitoring and support model for AI Gate.
Gateway first-response overheadp95 ≤ 100 msproposed enterprise SLA model per calendar month
Gateway availability≥ 99.9%controlled API gateway
Synthetic test3,000 RPSup to 180,000 RPM; concurrency up to 150,000
Token test1B TPMsynthetic test of the controlled gateway
Availability check60 secproposed external synthetic monitoring cadence
P1 first response≤ 1 hour24×7 support model
SLA figures describe the proposed enterprise technical/contractual model; a signed agreement takes precedence.
gateway_overhead_ms
Additional first-response delay introduced by the controlled gateway. TTFT is used for streaming and TTFB for non-streaming. External network, upstream queue/inference and payload transfer are excluded.
Observability
Request history includes duration_ms, upstream_first_token_ms, e2e_first_token_ms and gateway_overhead_ms; aggregated statistics use p50/p95/p99.
Availability methodology
Proposed model: external synthetic monitoring of /health/ready every 60 seconds with a 5-second timeout. Downtime is recorded after two consecutive failed checks. Client network and external upstream failures are excluded when the controlled gateway is healthy.