Vercel adds Browserbase web tools to AI Gateway and private npm installs to Vercel Agent

Vercel has shipped two developer-tooling updates for its AI platform: AI Gateway can now expose Browserbase's web search and page-fetching tools to any model that supports tool calling, and Vercel Agent can install private packages from npm and custom registries. Both changes were announced on the Vercel blog on October 1.

Browserbase tools arrive in AI Gateway

According to the Vercel blog, Browserbase Search and Fetch are now available through AI Gateway, sitting alongside Browserbase's existing integrations with the AI SDK and the Vercel Marketplace. Search lets a model find relevant, current pages on the web; Fetch retrieves the contents of a specific page. The two can be chained, for example by asking a model to locate a company's latest announcement, open it, and summarise it.

The notable part is the delivery mechanism. AI Gateway exposes these tools across model providers behind a single API key, so developers can attach web search and page retrieval to any tool-calling model and keep the same tool definitions when they switch models. Vercel describes the gateway as a unified interface for invoking models, tracking usage and spend, and configuring routing, retries and failover, with the option to bring your own provider key rather than relying on a gateway key.

On the technical side, the helpers ship in AI SDK 7.0.116 and later. Getting started means updating the ai package (for example with pnpm add ai@latest), setting an AI_GATEWAY_API_KEY, and registering browserbase_search and browserbase_fetch in the tools object of a call such as streamText. The search helper accepts options like a result count, while the fetch helper can be configured to follow redirects. Pricing and configuration details are covered in the Browserbase documentation, and the tools can be tried out in the AI Gateway model playground.

Vercel Agent installs private dependencies

The second update concerns Vercel Agent, the company's coding agent. Per the Vercel blog, Agent sessions can now install private dependencies from npm and custom registries, using credentials stored as shared environment variables on Vercel. The npm, pnpm and classic Yarn package managers authenticate inside Agent sessions the same way they already do in Vercel builds.

Setup requires adding a shared environment variable scoped to Development or Preview: NPM_TOKEN for private packages hosted on registry.npmjs.org, or NPM_RC to configure custom or multiple registries.

There is a deliberate security boundary in the design. Vercel Agent reads only team-shared variables, not project-scoped ones, and credential values stay outside the sandbox, which means the agent itself can never read the secrets its package installs rely on. Authentication happens at install time; the model simply does not get to see the token.

Why it matters

Both changes target friction in building and running AI agents. Provider-agnostic tools mean the tool layer does not have to be rebuilt whenever a team swaps in a cheaper or better model, and web search plus page retrieval are close to table-stakes capabilities for agent workloads that need current information.

Private package installs address a quieter but equally common blocker: coding agents are far less useful in repositories that depend on internal packages if they cannot authenticate against a private registry. Solving it with shared variables, while keeping credentials out of the agent's reach, mirrors how Vercel builds already handle the same problem. Taken together, the updates show Vercel investing in the infrastructure layer around agents rather than the models themselves.