NemoRouter runs on managed Cloud Run with multi-region failover, synthetic probes, and revision-scoped post-deploy log scans. Targeted uptime is 99.9% across every plan tier — the same SLA on Tier 1 as on Enterprise.
Failures rarely take down everything at once. Each surface is reported separately so partial degradations show up honestly — never papered over by a single green light.
Some systems are experiencing issues
As of May 19, 2026
API proxy — US-Central (primary)
OpenAI-compatible /v1 endpoint, intelligent router
Multi-region failover
Automatic failover when the primary degrades
Provider routing & fallback
Routing strategies, fallback chains, retry policies
Authentication
Login, signup, SSO/SAML, session management
RLS-scoped Supabase
Tenant-isolated Postgres with Row-Level Security
Virtual key validation
Per-key auth, RPM/TPM enforcement, budget caps
Credit ledger
Reserve + settle, atomic balance mutations
Stripe billing & webhooks
Checkout, top-ups, subscription, webhook ingest
Management dashboard
Keys, teams, analytics, guardrails, settings
Observability & logging
Request logs, callbacks, alerts, latency metrics
Anthropic
Claude Opus 4.8, Sonnet 4.6, Haiku 4.5 — live today
Google Vertex AI
Gemini, Imagen, Veo, embeddings — live today
OpenAI
GPT-5.5, GPT-4o, o-series — live today
AWS Bedrock
Shipping next — not yet routing production traffic
0 customer-impacting incidents this quarter (as of May 19, 2026). Major incidents (P0/P1) get a postmortem in the changelog within 7 days — timeline, root cause, remediation.
AWS Bedrock is marked “under maintenance” because it is shipping next — it does not yet route production traffic. Anthropic, Google, and OpenAI are live today. See the model catalog for what is live today.
Committed on every plan tier
Design target, multi-region
p50 added overhead, no extra hop
Customer-impacting; cosmetic excluded
Uptime is an output. The inputs are synthetic probes, deploy-time gates, and customer-configurable alerts that fire before a ticket gets filed.
Synthetic checks run against /v1/chat/completions and /health/readiness from every active region every 60 seconds — against the real proxy path with a virtual key, streaming and non-streaming separately. A miss alerts on-call before any customer ticket.
Every Cloud Run revision must pass a revision-scoped log scan before it serves traffic — boot logs scanned for DataError and 5xx for 60 seconds, Prisma drift blocking promotion, auto-rollback if the SLO window degrades. A 200 on /health is necessary, not sufficient.
Configure per-org alerts for budget burn, error spikes, latency regressions, or provider outages — 8 alert types with per-key and per-org thresholds, delivered to webhook, email, Slack, or Teams.
Not yet — and we will not pretend otherwise. This board reflects the operational posture as of the snapshot date and changes with the deploy that resolves any incident. A live, server-backed feed is on the roadmap. We chose not to embed a third-party vendor widget (Statuspage / Better Stack) because it requires a separate account and a public page that drifts from our deploys.
Synthetic probes hit the live API proxy from every active region every 60 seconds, and readiness gating blocks any revision that regresses before it serves traffic. We hold to a 99.9% availability target and count only customer-impacting incidents — the kind that surface in real production traffic; cosmetic and self-healing transient errors do not. Any P0/P1 gets a changelog postmortem.
Major incidents (P0 / P1) get a postmortem in the changelog within 7 days, with timeline, root cause, and remediation. As of the snapshot date there are no open or recent customer-impacting incidents.
Notice an issue?
If you are seeing routing errors or latency spikes that are not reflected here, our engineers want to know. Reports are triaged within an hour during business windows.
Security-sensitive reports — please email security@nemorouter.ai instead of filing a public issue.