Guides

Smart routing (rodiumai/smart)

Pass model: "rodiumai/smart" on POST /v1/chat/completions. An LLM classifier chooses the best catalogue model, then the gateway runs your completion on that model.

POSThttps://api.rodiumai.io/v1/chat/completions

How it works

Smart is a virtual model id, not a fixed upstream. Your request is classified, routed to a concrete catalogue slug (for example openai/gpt-4o or anthropic/claude-sonnet-4-6), then answered like a normal chat completion.

Two hops

Hop 1 classifies the prompt (router). Hop 2 runs the selected model. Both consume tokens and RODI; responses include rodiumai_routing metadata with the requested alias and the resolved model.

rodiumai_routing

Look for rodiumai_routing.requested (always rodiumai/smart) and rodiumai_routing.resolved (the catalogue model that answered). Use this for debugging and cost attribution.

Aliases (legacy)

Canonical id: rodiumai/smart. Still accepted: smart and rodium/smart. Prefer rodiumai/smart in new code; GET /v1/models lists it first.

Billing

You pay the router hop plus the selected model’s published rates. There is no separate flat Smart fee beyond those two inference hops.

When to use Smart

Ideal when prompts vary (coding, vision, short Q&A) and you do not want to hardcode a model. Pin a concrete slug when you need deterministic cost or behaviour.

Examples

…

Sample request

…

…

Same endpoint as chat: POST /v1/chat/completions · Chat completions guide