Route decision
Definition
A route decision is the explicit selection of a model, provider, or execution path for an AI request. It should include the eligible options, the policy inputs, and the selected destination so the outcome can be inspected later.
A decision is more than a destination
A route decision should say why a request went to one destination instead of another. Useful context includes the requested capability, model eligibility, policy version, recent provider health, budget, and the final destination. Without that context, an operator sees only a model name and cannot tell whether the gateway followed policy or fell through to a default.
Example
Imagine a coding request that requires tool calling and a large context. The gateway filters out models without those capabilities, compares the remaining options, and selects one with an acceptable latency and cost profile. The event records the constraints, the selected model, and the reason code. A later review can determine whether the choice met the request rather than rewarding a route merely because it returned HTTP 200.
Limitations
A recorded decision does not prove that the policy was wise or that the provider delivered good quality. Measure downstream completion, retries, user feedback, and total spend. Keep sensitive prompt content out of logs unless the application has a clear retention and access policy.
Weave context
For Weave Router, route decisions are most useful when joined to the coding task they support. Compare decisions across task types and model families, and preserve enough metadata to explain changes after a routing-policy update.
How this relates to Weave
Weave Router makes a model choice for coding-agent prompts. Treat the route decision as an observable event, then compare it with task success, latency, token use, and cost so the choice can be evaluated rather than assumed to be correct.
Explore Router