AI that keeps running when providers go down
TheRouter provides multi-provider failover for AI requests. When Anthropic, OpenAI, or Bedrock has an outage, TheRouter automatically routes to the next healthy provider. No code changes required. Health checks run continuously. Priority routing ensures the best provider is always used first, with automatic fallback.
TheRouter routes across Anthropic, OpenAI, Bedrock, Vertex AI, xAI, and SiliconFlow with real-time health checks. When a provider goes down, your requests automatically move to the next healthy route — your users never see provider outages.
Failover architecture
Your Application
│
▼
┌─────────────────────────────────┐
│ api.therouter.ai/v1 │
│ │
│ ┌──────────────────────────┐ │
│ │ Route Selector │ │
│ │ Priority 0: Bedrock │───┼──▶ AWS Bedrock (us-east-2)
│ │ Priority 1: SiliconFlow │───┼──▶ SiliconFlow (Aliyun)
│ │ Priority 0: Anthropic │───┼──▶ Anthropic API
│ │ Priority 0: OpenAI │───┼──▶ OpenAI API
│ │ Priority 0: Vertex AI │───┼──▶ Google Vertex AI
│ │ Priority 0: xAI │───┼──▶ xAI API
│ └──────────────────────────┘ │
│ │
│ Health checks run continuously │
│ Failed routes → skip + alert │
└─────────────────────────────────┘
│
▼
Normalized Response
(consistent model name,
finish_reason, structure)Priority-based routing
Each model has one or more provider routes with priority levels. Requests always go to the lowest-priority-number (highest-priority) healthy provider:
# deepseek/deepseek-r1 routes: # Priority 0 → AWS Bedrock us-east-2 (primary, US region) # Priority 1 → SiliconFlow cn (fallback, China region) # If Bedrock returns 503 or times out: # → TheRouter automatically tries SiliconFlow # → Response comes back with model: "deepseek/deepseek-r1" # → No error visible to your application # Same standard model name in response regardless of which # provider actually served the request
Common questions
How fast is failover when a provider goes down?
TheRouter detects provider failures in real time as requests are processed. The next request in the same window is automatically routed to the next healthy provider — typically under 100ms routing overhead.
What providers does TheRouter support for failover?
Currently: Anthropic, OpenAI, AWS Bedrock (us-east-2), Google Vertex AI, xAI, and SiliconFlow (China region). Each provider runs as an independent service — a failure in one does not affect others.
Do I need to handle retries in my application code?
No. TheRouter handles provider-level failover transparently. Your application receives either a successful response or a 5xx error only after all configured routes have been exhausted.
Can different API keys have different failover configurations?
Yes. In the dashboard you can set a provider preference per API key: Auto (tries all providers in priority order), US-optimized (US providers first), or China-optimized (SiliconFlow first for shared models).