# Availability & SLA

> Markdown mirror of https://router.one/sla for AI assistants and crawlers. Router One is an OpenAI-compatible LLM API gateway.
> Last updated: 2026-08-26

Router One is in the request path, so availability is the headline number. This page documents how it is measured, what fallback does, and what enterprise customers can lock in contractually.

## Availability evidence

### Measurement

Router One does not yet publish a platform-wide monthly uptime percentage. The availability evidence published today comes from real gateway traffic — the per-model panels described below — and from the hourly reachability probe on the status page, not from a synthetic multi-city probe schedule.

### Exclusions

Availability measurement excludes scheduled maintenance announced ≥ 48 hours in advance via changelog and email to active accounts.

### Where it is published

Per-incident detail is published on the status page linked from the footer, which retains public incident entries for at least 90 days; each per-model panel carries its own last-updated stamp. No monthly availability number is attached to changelog entries.

### Per-model evidence

Availability at the model level is published on each model's detail page under /models: successful-response share, terminal 429 rate, and terminal 5xx rate over rolling 7-day and 30-day windows, computed from real gateway traffic and shown only once a window holds at least 100 eligible requests from at least 5 independent callers. Those are operational aggregates, not a contractual SLA.

## Failover behavior

### Trigger

When a named-model request hits a retryable upstream error, such as an eligible 5xx or timeout, the gateway may retry it on the next healthy provider serving that same model.

### Fallback latency

Added latency depends on how the upstream fails and how quickly the next provider responds; there is no fixed latency guarantee. The dashboard records the final route and request metrics, not each internal failed attempt.

### Model family constraint

Provider retry keeps the exact requested model and only moves between healthy providers serving that model. Router One does not silently swap a named-model request to a different model.

## Status visibility

### Public status

Active incident state is published on the status page linked from the footer and emailed to addresses subscribed to status updates.

### In your trace

For every request, the dashboard shows the final upstream provider, latency, status, tokens, and cost. It does not expose every internal failed attempt individually.

## Enterprise SLA

### Default contracts

Annual enterprise contracts can include a written SLA with service credits for missed availability targets. Pay-as-you-go customers don't have a contractual SLA but get the same fallback infrastructure.

### Coverage

The SLA covers Router One's reachability and routing layer. Per-model availability is constrained by upstream providers — the SLA accounts for that by measuring against the routed result, not any single provider.

### How to sign

Contact sales via /contact to negotiate target, exclusions, and credit schedule.

## FAQ

### Where can I see live incident status?

The status page linked from the site footer publishes active incidents, an hourly API reachability probe, and retains public incident entries for at least 90 days. Email subscription details are offered from the same page.

### Does Router One offer SLA credits for pay-as-you-go users?

Contractual SLA with credits is included only in enterprise annual contracts. Pay-as-you-go users get the same routing and fallback behavior, but no written credit schedule.

### What counts as downtime in the SLA?

Downtime is measured against the final routed result reaching the customer. If an eligible provider retry completes successfully within the contract's availability criteria, the request is not downtime. Scheduled maintenance announced ≥ 48 hours in advance is also excluded.

## See also

- Methodology: https://router.one/methodology
- Smart routing: https://router.one/routing-methodology
- China latency benchmark: https://router.one/benchmarks/china-latency
- Trust hub: https://router.one/trust
- Changelog: https://router.one/blog/changelog
- Canonical page: https://router.one/sla
- Models and per-model token rates: https://router.one/models (markdown: https://router.one/models.md)
- Pricing: https://router.one/pricing
- API docs (markdown): https://router.one/docs.md
- Company facts: https://router.one/facts/company.md