Match the request.
Consider the work, the tools it requires, and the models in your allowed pool.
Solutions / Intelligent model routing
Match the work to an eligible model. Account for tools, context, cache state, and quality before making a switch.
Caveman Router + Cloud · routing in development
The model you trust becomes the model you use for everything.
Consider the work, the tools it requires, and the models in your allowed pool.
A lower token price can lose to cache rebuilding. Evaluate the switch in the context of the session.
Only compare qualified candidates. When a switch is unsupported, retaining the baseline is a valid result.
Price is part of the decision. Change the tool and cache requirements in this example to see why a candidate qualifies—or gets held back.
Turn the constraints on and off. Watch which route remains eligible.
Small model excluded: required tool capability is missing.
Illustrative routing policy. Real decisions depend on configured capabilities, quality evidence, prices, context limits, and session economics.
{
query: string,
limit: number
}Arguments validatedA candidate must support the required tool and output formats before it gets ranked.
Preserve valuable cached context when a switch does not justify rebuilding it.
Keep quality and policy checks between a candidate route and production traffic.
Inspect each stage. Every promotion has a purpose, and production keeps a baseline to return to.
Capture representative traffic before proposing a change.
Stage previewRouter makes the choice. Platform keeps the cost, usage, and routing record in view. A route report should explain the change—including a decision to leave traffic where it was.
Treat the complete model action as the route identity.
Keep the comparison explicit, with an explanation for rejected alternatives.
Keep modeled opportunity separate from provider-causal savings.
Make every token count.
Go deeper into cache-aware routing, capability gates, and the economics of a completed task.