OpenAI has updated its agents software development toolkit (SDK) with new features aimed at helping businesses create automated assistants. The new version includes sandboxing abilities that allow agents to operate in controlled environments, reducing risk from unpredictable behavior. Additionally, developers now have an in-distribution harness for frontier models, enabling agents to work with files and approved tools within a workspace.
The update is designed to make it easier for companies to deploy and test long-horizon agents on OpenAI's models. Long-horizon tasks involve complex, multi-step work that requires advanced capabilities. "This launch is about making our existing Agents SDK compatible with all sandbox providers," said Karan Sharma, an OpenAI product team member.
The new features are initially available in Python, with TypeScript support planned for a later release. OpenAI plans to expand the Agents SDK over time, including adding code mode and subagents capabilities to both languages.
OpenAI's updated Agents SDK is being offered to all customers via the API and will use standard pricing. The company aims to help enterprises create their own agents by providing tools that are compatible with a range of sandbox providers.
