Vercel now supports running Claude Managed Agents inside Vercel Sandbox. Each agent session runs in an isolated Firecracker microVM, giving developers a secure execution environment for Claude's tool calls against private APIs and internal services. Key features include per-session VM isolation with millisecond startup, credential brokering at the firewall so secrets never enter the sandbox, deny-by-default egress with a domain allowlist, and low-latency connectivity to private networks. Anthropic handles the model, harness, tools, and session state, while Vercel provides the execution infrastructure.
Sort: