What goes wrong
Rename a helper. Explain a short error. Then a real design review. From the editor, those often look like the same kind of request. They are not the same cost — but they still leave toward the same expensive model.
The usual fix is to reconfigure the editor: swap provider, change model, fight URLs and API shapes. That works once. It gets brittle when you want a cheap local model for easy turns and a cloud model for hard ones — without rebuilding settings every time.
Why it matters
Model choice is not only a preference in a settings panel. It is a traffic decision. If the policy lives only in the client, you keep paying for a one-size-fits-all default — and switching providers means reconfiguring every tool that talks to them.
What good looks like
Keep the editor on its normal path. Put the policy on the requests themselves: send hard work to the cloud and easy work local; switch providers when you need to — without the editor knowing the destination changed.
Test scenario · Token threshold
Large codebase review
A long code-review prompt. A simple rule changes the destination before the provider sees the request — the editor still asked for the original model.
gpt-4o gpt-4o-mini Your editor still requests the original model. Aura Router rewrites host, path, and provider transparently — including cross-provider translation with streaming — and you see the swap in Live Traffic and Working samples.
How Aura Wire helps
Aura Wire sits on the path between your editors and AI providers. Rules can change where a request goes — without storing prompts or answers.
- Start in suggest mode — see recommended routes from real traffic before anything rewrites
- Send hard work to the cloud, easy work to a local model, or let a request fall through
- Talk to different providers without rebuilding each editor's settings
- See estimated savings when rules divert traffic to cheaper or local models
Try Aura Wire free
AI chat for VS Code with a local pipeline behind it — live spend, secrets stopped, nothing stored.