OpenAI Ships Agents SDK With Native Sandbox Execution — Seven Provider Integrations for Isolated Agent Code Runs

OpenAI·medium signal

OpenAI released a major Agents SDK update on April 15-16 adding native sandbox execution, so agents can run shell commands, write files, install packages, and use tools inside isolated environments without touching the host machine. Built-in sandbox support ships for Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, and Vercel — or bring your own. The update also includes configurable memory, sandbox-aware orchestration, Codex-like filesystem tools, and standardized integrations. Python first, TypeScript planned. This moves agent sandboxing from a DIY infrastructure problem to a first-party SDK feature.