The market view your GPU spend is missing.
Cloud GPU pricing is negotiated in the dark: every provider quotes alone, and the spot market that undercuts them is scattered across marketplaces nobody has time to watch. This site is the whole tape in one place — live, historical, and machine-readable.
Price discovery, cross-provider.
One ranked table normalizes marketplace hosts and cloud price lists onto the same number: dollars per GPU per hour. Before a reserved-contract renewal, thirty seconds here tells you what the open market charges for the same silicon.
start: the live market · /api/offers
History that survives the negotiation.
We record every provider's best and median price per GPU model, every poll, permanently. When a vendor says "that price was an anomaly", the tape answers — charted on the site, raw over the API.
start: the tape · /api/history
Alerts under your budget line.
Arm a watch at the price your finance model assumes. When the market crosses under it, your Slack, RSS reader, or internal webhook rings — no account, no vendor lock, nothing installed.
start: the tripwire
Your agents are already welcome.
Platform and FinOps teams wiring internal AI agents get the same data as native MCP tools — search, history, the idle index, watches — listed on the official MCP Registry. No key, no contract, no rate negotiation.
start: the integration guide
Why it costs nothing.
The platform fee is zero and every response says so (fee_bps: 0). Outbound provider links may carry referral tags the provider pays for; your price is unchanged. That's the whole business model, stated in one sentence.
On the bench, not yet shipped.
Brokered execution — placing rentals on your own provider accounts through one API — is designed and in build. It ships when it passes real-money tests, not before. Nothing on this page describes a feature that doesn't exist today.