Run untrusted code with Vercel Sandbox, now generally available
Not a pilot. Vercel Sandbox now runs AI agents at scale—2.7M deployments/day infrastructure, open-sourced, sub-second cold starts.
Why it matters
AI agents need fundamentally different infrastructure than human developers. Vercel's GA launch of Sandbox—purpose-built for ephemeral, isolated, untrusted code execution—removes a critical blocker for agent-native software workflows and shifts how infrastructure vendors compete for agent workloads.
The key facts
8 to knowVercel Sandbox now generally available
Infrastructure spans 2.7M deployments/day at production scale
Sub-second cold starts for sandbox environments
Full isolation + sudo access + package managers for untrusted code
Ephemeral design with Active CPU pricing (pay for execution, not idle)
Open-sourced CLI and SDK for community adoption
Early adopters: Roo Code (multi-platform AI coding agents), Snapshots (environment restoration), Blackbox AI (multi-agent orchestration platform with horizontal scaling)
Powered by Firecracker microVMs orchestrated across multi-region clusters (internal 'Hive' platform)
Go to the source
Vercel Blogvercel.com
Publisher excerpt: AI agents are changing how software gets built. They clone repos, install dependencies, run tests, and iterate in seconds. Despite the change in software, most infrastructure was built for humans, not agents. Traditional compute assumes someone is in the loop, with minutes to provision and…