Routing

Sylica routing extensions let you shape provider behavior directly when latency, cost, or policy constraints are strict.

By default, provider selection is automatic and balanced for stability. For specialized workloads, pass an explicit provider policy block to constrain or prioritize providers without changing your endpoint contract.

Routing Fields

Use provider.order to express preference order, provider.require to hard-limit which providers may execute, and provider.allow_fallbacks to control whether fallback attempts are allowed before first-byte emission.

Catalog Response Sample

json

{
  "object": "list",
  "data": [
    {
      "id": "openai/gpt-5-mini",
      "provider": "openai",
      "display_name": "GPT-5 mini",
      "context_window": 400000,
      "pricing": {
        "input_per_mtok_usd": 0.25,
        "output_per_mtok_usd": 2.0
      },
      "capabilities": {
        "vision": true,
        "tools": true,
        "reasoning": true
      }
    }
  ]
}

Latency Strategy

For chat UI, route to low-latency models first and stream aggressively. For back-office summarization or offline transforms, use lower-cost paths and permit fallbacks more freely. Keep these policies separate by workload class rather than trying one policy for every request type.