Vercel's AI Gateway now lets you set hard spend caps on individual API keys. Once a key hits its dollar limit, the gateway rejects all further requests until the budget resets or you raise it manually. The cap covers every provider and model routed through that key.
Budgets can be created from the AI Gateway API Keys dashboard or via the Vercel CLI. Refresh periods are configurable: daily, weekly, monthly, or none, with each window resetting at UTC start. Keys provisioned through the CLI appear in the dashboard automatically, so there is one consolidated view regardless of how they were created.
The targeting logic is what makes this worth reading in full. Vercel calls out three specific failure modes this addresses: agentic loops that fan out without supervision, prototypes that get unexpected traffic after being shared, and developers experimenting without per-model cost visibility. The budget mechanism is a control surface for exactly those cases, not just a billing convenience.
[READ ORIGINAL →]