Vercel AI Gateway adds per-user budgets to cap runaway spend

The gateway now meters each teammate against a dollar cap; Vercel also drops an fx AI SDK adapter and daily sandbox storage billing. Plus Debian's AI vote.

Nowline SEP 1 9:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Cap what each teammate can spend

    AI Gateway budgets now scope to individual users, on top of team, project, and API-key limits. Set a dollar cap per member and the gateway halts their requests once it's hit — a hard guardrail against one runaway agent or one teammate draining the shared balance.

  • fx lands in the AI SDK harness

    A new @ai-sdk/harness-fx adapter wires fx into the AI SDK harness layer through the HarnessAgent interface, built on the ACP harness. If you build agents on the AI SDK, it's a fresh drop-in path to run them.

  • Sandbox snapshot storage now billed daily

    Vercel Sandbox now recalculates snapshot storage costs every day at $0.08 per GB-month, so charges track what you actually keep rather than a coarse estimate. If your agent runs stash sandbox snapshots, watch the new line item.

  • Debian says yes to AI-assisted code

    Debian developers passed a General Resolution permitting 'responsible' use of generative AI in contributions — human review and testing mandatory, and the contributor fully accountable for every line. One of open source's most conservative projects just opened its doors to AI-written patches.

  • 25 GPU clouds audited, many leaking

    SemiAnalysis spent four months probing 25 GPU cloud providers and found widespread lapses: cross-tenant remote-code-execution from years-old CVEs, misconfigured InfiniBand keys, and exposed monitoring dashboards. Renting from a smaller 'neocloud'? Don't assume tenant isolation holds.