OpenAI Agents SDK: Native Sandbox Execution with Manifest Abstraction — Mount Local Files, Define Outputs, Swap Providers (E2B, Modal, Vercel, Cloudflare)

OpenAI·medium signal

OpenAI's Agents SDK now supports sandbox execution as a first-class primitive. The Manifest abstraction describes an agent's workspace — mount local files, define output directories, and connect storage providers (S3, GCS, Azure Blob, R2). Supported sandbox providers: Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, Vercel. The model-native harness keeps planning, tool selection, and self-correction inside the reasoning chain, reducing malformed tool calls. Python first, TypeScript later.