Vercel

Vercel AI Gateway now supports per-user spend limits, covering every API key and app token attributed to a person. Budgets stack with key, project and team limits, reset on a daily, weekly or monthly cadence, and alert by email at 50, 75 and 100 percent of allocation.

An agent framework adds an iMessage channel: each conversation maps to one session, webhooks are verified, and messages arriving mid-reply steer the in-progress turn rather than spawning separate answers.

Vercel opens early access to Token Pass for Pro teams: $80 monthly credit for GLM 5.3 and Kimi K3 at $30, routed through AI Gateway with zero data retention, usable from coding agents or the AI SDK.

Run SDK executes untrusted agent-written JavaScript and TypeScript in a fresh QuickJS context inside a worker thread, with no direct path to Node or the network. Applications expose narrow host functions as sandbox globals and can durably interrupt a program for auth or human approval, resuming afte

The Vercel CLI gained a feedback command for sending bugs and feature requests from the terminal. Agents can attach structured key=value context such as model, harness, task, expected and actual behavior, or pipe in notes from a file.

Chat SDK picks up an XChat adapter: bots hold end-to-end encrypted direct and group conversations with encryption, keys and signature verification handled by the adapter. Since XChat renders no markdown, tables degrade to ASCII blocks and streaming arrives as message edits.

Vercel Connect adds Enterprise-Managed Authorization in public beta, supporting Cross-App Access for MCP. Admins control agent access via the identity provider and Connect handles token exchange and renewal.

Vercel Connect adds a Connector Permissions setting: Pro and Enterprise team owners can now restrict connector creation and management to Owners or a new Connector Manager role, tightening control over agent/app access to external credentials.

Vercel CLI adds a searchable changelog command so coding agents can query recent product updates and feature announcements directly from the terminal instead of scraping the web.

Vercel Daemons enters public beta, running containerized long-lived workers that poll queues or task systems. Vercel builds the image, keeps the configured replica count per region and restarts crashed instances.

Cursor's Self-Hosted Machines API now has a Vercel Sandbox reference implementation: each agent request gets its own Firecracker microVM, while Functions and Workflow claim queued requests, provision workers, monitor sessions and clean up. Credentials inside each sandbox are short-lived and user-sco

Exa lands on Vercel as a native integration. Pass gateway.tools.exaSearch() alongside a prompt and the gateway executes the searches, feeds results back to the model, and returns one synthesized answer — with the usual Exa controls for result count, category, country and domain filters, billed next

Vercel's AI SDK harness layer now wraps the Agent Client Protocol itself rather than individual runtimes, so any ACP-compatible harness can be adapted by passing its package to createACP.

Vercel reports the outcome of its sandbox escape challenge: 1,285 reports, $242,000 in bounties, and a Linux kernel networking zero-day that let a microVM guest leak host memory, crash the host or execute code. No customer data access was found.

Alibaba's Wan 3.0 arrives on AI Gateway as a single model spanning text-to-video, image-to-video, first/last-frame conditioning and reference-based generation, taking image, video and audio references. Up to 30s at 30fps and 1080p with audio by default, paired with the new async generation paths.

Vercel's AI SDK harness layer now supports GitHub Copilot via an official ACP-based adapter, joining Claude Code, Cursor, Codex, Cline and others behind one HarnessAgent interface so apps can swap coding agents without code changes.

Vercel Agent now authenticates npm, pnpm and classic Yarn against private registries using team-shared NPM_TOKEN or NPM_RC variables, with credential values held outside the sandbox so the agent cannot read them.

Vercel's eve agent framework adds persistent memory slots scoped by provider (file, Supermemory, Upstash, Kybernesis) so agents retain user-specific context across sessions and deployments.

Vercel Sandbox now runs Harbor, the harness behind Terminal-Bench and other agent benchmarks like SWE-bench and OSWorld, in isolated Firecracker microVMs, letting engineers parallelize evals across models via a single AI Gateway key.

Vercel explains how its Agent investigates under read-only access, then proposes a plan whose extra scopes are approved as a single decision, avoiding both broad standing permissions and approval-click fatigue.

Vercel now lets any individual deployment opt into faster Turbo build machines via a git commit tag, the CLI's --turbo flag, or the REST API, instead of a project-wide setting change.