Routing & Failover

Cencori's Routing Engine allows you to decouple your application logic from specific providers.

You can configure routing rules in the Dashboard > Gateway > Policies.

• Lowest Cost: Routes to the cheapest provider that meets the model capabilities (e.g., gpt-4o vs claude-sonnet-4.5).
• Lowest Latency: Routes to the provider with the fastest current response time.
• Performance: Routes based on historical quality scores (user feedback).

You can configure project-wide failover settings directly in the dashboard:

1. Go to Settings > Providers.
2. Scroll to the Fallback Configuration section.
3. Toggle Enable automatic fallback.
4. Select a Fallback Provider (e.g., if OpenAI fails, switch to Anthropic).
5. Set Max Retries (default is 3 attempts before switching).

For more complex logic, you can define routing policies via the Gateway API:

Example Configuration:

// Policy: "Reliability First"
{
  "primary": "openai/gpt-4o",
  "fallbacks": [
    "anthropic/claude-sonnet-4.5",
    "google/gemini-1.5-pro",
    "mistral/large"
  ],
  "conditions": {
    "on_status": [429, 500, 503, 502]
  }
}

When OpenAI returns a 503 Service Unavailable, Cencori instantly rewrites the request and sends it to Anthropic.

Distribute traffic across multiple API keys organization-wide to avoid hitting rate limits on a single account.

• Round Robin: Cycles through keys evenly.
• Weighted: Assign more traffic to higher-tier accounts.
• Rate Limit Aware: Automatically shifts traffic away from keys nearing their limit.