Skip to main content
ARouter continuously monitors the health and availability of all upstream AI providers to ensure maximum uptime for your applications.

​
How It Works

ARouter tracks response times, error rates, and availability across all providers in real time. This data drives intelligent routing decisions and helps surface reliability information in your Activity feed. When a provider experiences degraded performance or an outage, ARouter automatically adjusts routing weights to deprioritize that provider — without any change required on your side.

​
What ARouter Monitors

For each provider and model, ARouter continuously tracks:
  • Success rate: Percentage of requests that complete without error
  • Time to first token (TTFT): Latency from request submission to first streaming token
  • Total response time: End-to-end latency for non-streaming responses
  • Error types: Distinguishes between transient errors (5xx, rate limits) and permanent errors (invalid model, bad request)

​
Automatic Routing Around Outages

When ARouter detects that a provider is degraded:
  1. The provider’s routing weight is reduced or zeroed temporarily
  2. Subsequent requests are routed to other healthy providers serving the same model family
  3. The provider is re-evaluated periodically and reintroduced once health metrics recover
This happens transparently — your requests continue to succeed even during provider incidents.

​
Customizing for Higher Availability

​
Use Ordered Candidate Model Lists

For critical workloads, specify an ordered list of models. ARouter tries each in sequence until one succeeds: 
{
  "models": [
    "anthropic/claude-sonnet-4-6",
    "openai/gpt-5.4",
    "google/gemini-2.5-pro"
  ],
  "route": "fallback"
}
This ensures your application keeps working even if the primary model’s providers are fully unavailable. See Model Routing for the full configuration reference.

​
Use Auto Routing

Set model: "auto" to let ARouter dynamically select the best available model based on current provider health, cost, and capability: 
{ "model": "auto" }
See Model Routing — Auto Routing.

​
Use :floor for Cost-Stable Routing

The :floor suffix routes to the lowest-cost provider serving a model, which is often a different provider than the default — providing natural diversity: 
{ "model": "anthropic/claude-sonnet-4-6:floor" }

​
Monitoring Provider Health

View real-time provider health data in the ARouter Dashboard. The dashboard shows per-model availability and latency trends so you can make informed decisions about model selection for your workloads.