OpenAI introduced new capabilities to the Agents SDK on April 15, 2026, enabling developers to build agents that can inspect files, run commands, edit code, and work on long-horizon tasks within controlled sandbox environments. The updated Agents SDK provides standardized infrastructure designed specifically for OpenAI models, combining a model-native harness with native sandbox execution for safe operation.
The new harness allows agents to work across files and tools on a computer while maintaining security through isolated execution environments. Developers can now give agents a controlled workspace, explicit instructions, and the tools needed to complete complex tasks. The SDK supports configurable memory, sandbox-aware orchestration, and filesystem tools similar to Codex functionality.
Enhanced Harness Capabilities for Agent Operations
The Agents SDK harness now includes several standardized integrations becoming common in frontier agent systems. These include tool use via MCP, progressive disclosure via skills, custom instructions via AGENTS.md, code execution using the shell tool, and file edits using the apply patch tool. The harness will continue to incorporate new agentic patterns over time, allowing developers to focus on domain-specific logic rather than core infrastructure updates.
According to Rachael Burns, Staff Engineer and AI Tech Lead at Oscar Health, the updated SDK made it production-viable to automate critical clinical records workflows. She stated, “The difference was not just extracting the right metadata, but correctly understanding the boundaries of each encounter in long, complex records.” This capability helps healthcare providers more quickly understand patient information during visits.
“The updated Agents SDK made it production-viable for us to automate a critical clinical records workflow that previous approaches couldn’t handle reliably enough.”
Rachael Burns, Staff Engineer & AI Tech Lead, Oscar Health
Native Sandbox Execution and Environment Management
The updated Agents SDK supports sandbox execution natively, allowing agents to run in controlled computer environments with the files, tools, and dependencies needed for tasks. Developers can bring their own sandbox or use built-in support for Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, and Vercel. The SDK introduces a Manifest abstraction for describing the agent’s workspace, enabling developers to mount local files, define output directories, and bring in data from storage providers including AWS S3, Google Cloud Storage, Azure Blob Storage, and Cloudflare R2.
This approach gives the model a predictable workspace where to find inputs, write outputs, and keep work organized across long-running tasks. The separation of harness from compute improves security by keeping credentials out of environments where model-generated code executes. It also enables durable execution through built-in snapshotting and rehydration, allowing the SDK to restore agent state in a fresh container if the original environment fails.
Security, Durability, and Scalability Design
Agent systems designed with separated harness and compute architecture address security concerns including prompt injection and exfiltration attempts. When the agent’s state is externalized, losing a sandbox container does not mean losing the entire run. Agent runs can use one sandbox or many, invoke sandboxes only when needed, route subagents to isolated environments, and parallelize work across containers for faster execution. This design makes agents more secure and scalable for production deployments.
Availability and Future Development
The new Agents SDK capabilities are generally available to all customers via the API using standard API pricing based on tokens and tool use. The new harness and sandbox capabilities are launching first in Python, with TypeScript support planned for a future release. OpenAI is also working to bring additional agent capabilities, including code mode and subagents, to both Python and TypeScript.
The company plans to expand the agent ecosystem over time with support for more sandbox providers, additional integrations, and more ways for developers to plug the SDK into tools and systems they already use. This approach aims to help developers bring more capable agents into production with less custom infrastructure while preserving the flexibility and control needed to fit agents into their own environments.
