manifesto

Neutral. By design.

We built whatmodel because the bill people pay for AI is not the bill they should be paying. Agents fan out across models, retry, re-prompt, and quietly burn the most expensive model on the most boring work. Nobody sees it. Nobody owns it. Until the invoice lands and the founder finds out.

So we made a managed service. Not a dashboard you wire yourself. Not a gateway you have to maintain. Not an SDK you have to wrap your code in. A service. You point us at a stack, and we run the cost ops from there — analyze every call, find the cheaper equivalent that does the same job, flag what repeats and could be cached, watch for drift, and tell you exactly what to change. The savings are yours to keep.

What we will not do

We will not pick a side.We do not sell a model. We do not sell a frontier lab's rebadged stack. Every recommendation we make is a real comparison against the live price index, on the day you asked. When a new model gets cheaper, you get the cheaper model. When an old model gets retired, you get the next-best equivalent. The only side we are on is yours.

We will not touch your keys. You connect once with the smallest credential that does the job — a provider admin key, read-only, scoped. Your traffic stays on your account, billed to your card, with your identity. We never see content, never store prompts, never proxy the bytes unless you turn on Optimize and tell us to. The audit alone needs no key at all — the keyless scan reads what your agents already wrote to disk.

We will not leak the recipe. The specific routing rule we apply on your behalf — which model, what cache key, which prompt slice we compressed — stays inside your workspace. Public pages show categories and totals, not the play. The recipe is yours. It is part of the reason the savings exist.

We will not take a cut of your savings. The audit is free. Always. After that you pay one flat monthly fee — sized only to how much spend you route through us, nothing more. Every dollar we save you, you keep. All of it. The number on your dashboard is proof of the work, not a meter we bill against. The more we save you, the more you keep. That is the whole point.

What we believe

The bill is a build problem. The team that ships fastest is the team whose costs are legible, attributable, and acted on automatically — not the team running the smallest model. The right model for the job is whichever one does the job for the lowest price today, and that is a moving target. Watching that target is a full-time job. It should not be yours.

Run the agents you want to run. Build the loops you want to build. Ship the product. We will price the calls.

See it on a demo workspace →← back to whatmodel