Vercel Brings Full Sandbox egress firewall to Its Hobby Plan.

Vercel is giving free-tier developers tighter control over what their sandboxed code can reach. The company’s Full Sandbox egress firewall, previously limited to Pro and Enterprise customers, is now available on the Hobby plan.

The change brings production-style network isolation to individual developers experimenting with AI agents, generated code and other potentially unpredictable workloads. According to Vercel’s official changelog, Hobby users can now restrict outbound connections, define approved destinations and keep authentication secrets out of the code running inside a sandbox.

That last capability matters. Sandboxes are designed to execute code in isolation, but many useful workloads still need access to external APIs, databases or internal services. Giving the code direct access to API keys creates another potential route for credential exposure. Vercel’s firewall can instead attach credentials to an outbound request after it leaves the sandboxed process, allowing the application to call an authenticated service without seeing the underlying token.

Vercel Sandbox users can configure the firewall by passing a networkPolicy when creating a sandbox. Policies can use one of three broad approaches:

• Allow all, permitting outbound connections by default
• Deny all, blocking outbound network access
• Custom rules, allowing selected domains or IP addresses

Custom policies can be more granular than a basic destination allowlist. Developers can use matchers to scope a rule according to the request path, HTTP method, query string or headers. This makes it possible to approve a specific API operation without granting unrestricted access to every endpoint hosted on the same domain.

The firewall also supports request transformations. Vercel’s example permits traffic to its AI Gateway domain, then adds an Authorization header containing a token stored in the surrounding environment. The sandboxed code makes the request, but the network layer supplies the credential.

This separation can reduce the damage caused by a malicious dependency, prompt-injection attack or generated script that attempts to inspect environment secrets. It does not make untrusted execution risk-free, but it narrows the routes through which code can communicate with outside systems or exfiltrate data.

The expansion is particularly relevant to developers building agents that run commands or generate and execute code. Those systems may need legitimate access to a small set of APIs, while having no reason to communicate with arbitrary internet destinations.

A developer could, for example, deny outbound traffic by default, allow requests to a model gateway and proxy selected requests through infrastructure they control. Credentials could be inserted only for matching destinations and request patterns. If executed code attempted to send data elsewhere, the firewall policy could block the connection.

The release also fits Vercel’s broader effort to position Sandbox as infrastructure for agent workloads. It follows the addition of Sandbox and AI Gateway support to Hermes Agent, as previously reported. Making the complete firewall feature set available on Hobby lowers the barrier to testing those architectures before a project moves to a paid plan.

For small teams, the immediate benefit is less about formal compliance and more about safer defaults. Builders can prototype applications with real external services without placing every credential inside the execution environment. They can also begin with restrictive network rules, then deliberately expand access as a workflow develops.

No separate product or upgrade is required for Hobby users. Existing Vercel Sandbox projects can adopt the capability by defining a network policy when creating a sandbox.

What is Vercel Full Sandbox egress firewall?

It is a set of network controls for Vercel Sandbox that determines which outbound connections sandboxed code can make. It supports allow-all, deny-all and custom policies based on domains, IP addresses and request attributes, while also allowing credentials to be attached outside the sandboxed process.

Which Vercel plans include the firewall?

All Vercel Sandbox firewall features are available on the Hobby plan as of August 5, 2026. The feature was previously available only to Pro and Enterprise customers.

How does a developer enable it?

A developer passes a networkPolicy configuration when creating a Vercel Sandbox. The policy can allow or deny traffic broadly, or define destination rules and matchers for paths, methods, query strings and headers.