Vercel enables secure AI code execution for Hermes Agent

Vercel has introduced new integrations for the Hermes Agent, a coding agent developed by Nous Research. The updates allow the agent to utilize the Vercel AI Gateway for model inference and Vercel Sandbox for secure code execution.

Through the AI Gateway, Hermes Agent can access over 200 different models. This centralized approach allows for live updates on model availability and pricing directly through the gateway dashboard.

To address security concerns regarding AI-generated code, Vercel Sandbox provides an isolated execution environment. Unlike standard containers, it utilizes Firecracker microVMs to create an ephemeral, compute-bounded boundary. This ensures that if an agent generates a malicious or erroneous command, the impact is contained within the disposable microVM rather than the developer's local machine or the application's primary process.

By default, Vercel Sandbox execution is disabled and must be explicitly configured. When enabled, the command execution environment shifts from the local workspace to a cloud-based microVM. While this increases security by isolating the filesystem and environment variables, users must manage local dependencies and authentication via Vercel OIDC tokens.