The financial layer of AI.
See, control, and prove every dollar your AI spends.
A live X-ray of your AI spend
Every request metered across providers, attributed per key, project, and route: the bill explained while it happens, not at month end.
Hard caps and an instant kill-switch
Budget alerts at 50, 80, and 95%. Caps that act, not just notify. And the gateway is designed to fail open: if we degrade, your traffic goes straight to the provider — everywhere except the cap you armed.
Receipts for every request
A per-request receipt and a hash-chained ledger you can re-derive offline. Numbers your finance team can check without trusting us.
Point Claude Code at us: see it and cap it.
Your agent's spend, live on a dashboard, with a hard cap behind it. Cache-level savings ship the moment they are measured, not before.
The referee that takes no money from the players.
BYO-key. No token markup. No provider money. The subject of the proof never pays for the verdict. That neutrality is the product.
We publish what we can prove. Here is what we can't, yet.
- Verified savings and gain-share are OFF: "proof pending." Until the evaluation gate is calibrated on real traffic, every savings figure reads measured, never verified.
- No savings claims on streaming or agentic traffic beyond the cache lever. We don't quote numbers we haven't measured.
- Routes default to observe: optimization levers are built but not yet activatable, and when they are it will be with your sign-off, never silently.
- No customer counts, no naked percentages: when a number appears on this site, its method is one click away.
The platform
The meter, the breaker, the ledger: spend infrastructure in the request path.
Done for you
The AI Spend Assessment: one fixed fee, priced by your spend band, delivered as a written report, with your dashboard live on the data behind it.
The program
The open methodology behind every number, and the open problem the program exists to solve: automated, provable evaluation.