OpenAI launches AgentKit to make building AI agents faster and safer: Details

OpenAI has introduced AgentKit, a groundbreaking suite of tools aimed at simplifying the process of building and managing AI agents for developers and organizations. This announcement was made on October 6, 2025, coinciding with the celebration of Developers’ Day. Previously, the development of AI agents posed significant challenges, requiring developers to navigate multiple tools, write extensive custom code, and invest considerable time into interface design. With AgentKit, these complexities are significantly reduced, allowing users to design workflows and create chat experiences more efficiently. At the heart of this new toolkit is the Agent Builder, a visual interface that enables developers to drag and drop components to construct multi-agent workflows. This feature enhances collaboration by allowing teams to visualize agent operations, test them, and implement changes with ease. Additionally, OpenAI has introduced the Connector Registry, which streamlines data management across various applications and workspaces, including popular platforms like Dropbox, Google Drive, Microsoft Teams, and SharePoint. To ensure the safety of these agents, OpenAI has incorporated Guardrails, a robust security feature designed to prevent harmful actions and safeguard sensitive information. For developers looking to incorporate chat functionalities, ChatKit simplifies the integration of chat-based agents into websites or applications, facilitating their use for customer support, onboarding, research, and knowledge assistance. AgentKit also enhances the testing process for developers with new features in Evals, allowing for automatic grading of agents, comprehensive workflow checks, and model testing from other companies. Major organizations, including Carlyle and Box, have reported improvements in development speed and accuracy thanks to these tools. Moreover, OpenAI is expanding its reinforcement fine-tuning (RFT) capabilities, enabling developers to train models to make improved decisions and adhere to customized rules. This feature is currently available for select models and is in beta for GPT-5. While ChatKit and Evals are immediately accessible to all developers, Agent Builder and Connector Registry are currently in beta. All tools are available under standard API pricing, with OpenAI promising additional features, such as a Workflows API and new deployment options, in the near future.