Building a production-grade, secure web-based AI agent that performs real work—writing and executing code, operating browsers, and orchestrating multi-step agentic workflows within a Vultr sandbox control layer. Containment-first and execution-driven.