Filter routes
support/triage
claude-sonnet-4 · 402k/24h
copilot/complete
gpt-4o · 377k/24h
data/extract
gemini-2.5-pro · 118k/24h
support/summarize
gpt-4o-mini · 296k/24h
chat/assistant
claude-sonnet-4 · 44k/24h
agents/plan
claude-opus-4.7 · 2.1k/24h
experimental/translate
draft · unused
support/triage
healthy
402k req / 24h · $1,020 · p95 1.84s
ENTRY
POST /v1/triage
any sk_live_prod_*
GUARDRAILS
PII redact · tox
block on fail
TRANSFORM
prompt pack
tools: 6 of 37
PRIMARY · 92%
claude-sonnet-4
Anthropic · cache on
FALLBACK · 7%
gpt-4o
on 429 · 5xx
FALLBACK · 1%
gemini-2.5-pro
on timeout
RESPONSE
stream back
log + enrich
timeout 30s
retry 2× / 500ms backoff
rate 2k rpm · team=support
cache ttl 10m
auto-fallback: on
A / B splitrunning 3 days
claude-sonnet-4 · 70%$0.183 · 1.84s
claude-sonnet-4-5 · 30% (canary)$0.171 · 1.52s
Canary wins on cost and p95 — promote?
Health · 24hper minute
deploy v12