News

OpenAI Enhances Agents SDK with Sandbox and Harness for Safer, More Capable Enterprise AI Agents

OpenAI Enhances Agents SDK with Sandbox and Harness for Safer, More Capable Enterprise AI Agents

Agentic AI is emerging as a significant success story in the tech industry, with companies like OpenAI actively providing enterprises with tools to create these automated assistants. To this end, OpenAI has updated its Agents Software Development Toolkit (SDK), introducing new features designed to help businesses build safer, more capable AI agents powered by OpenAI's models.

A key new capability in the SDK is a sandboxing ability, which allows agents to operate within controlled computer environments. This is crucial because agents can sometimes be unpredictable, making unsupervised operation risky. With sandbox integration, agents can function in a siloed capacity within a specific workspace, accessing files and code only for particular operations while protecting the system's overall integrity.

OpenAI also stated that the new SDK version provides developers with an in-distribution harness for frontier models, enabling these agents to work with files and approved tools within a workspace. In agent development, a "harness" refers to an agent's components beyond the core model it runs on. An in-distribution harness typically facilitates both the deployment and testing of agents utilizing frontier models, which are considered the most advanced general-purpose models available.

Karan Sharma from OpenAI's product team explained, "This launch, at its core, is about taking our existing Agents SDK and making it so it’s compatible with all of these sandbox providers." He hopes that this, combined with the new harness capabilities, will enable users "to go build these long-horizon agents using our harness and with whatever infrastructure they have."

Long-horizon tasks are generally defined as more complex and multi-step workflows. OpenAI plans to continue expanding the Agents SDK over time. Initially, the new harness and sandbox capabilities are launching in Python, with TypeScript support planned for a later release. The company is also working to bring additional agent capabilities, such as code mode and subagents, to both Python and TypeScript.

The new Agents SDK capabilities are available to all customers via the API, utilizing standard pricing.

↗ Read original source