OpenAI has unveiled AgentKit, a new suite of tools designed to make building and managing AI agents faster, safer, and more intuitive. The announcement, made on October 6, 2025, coincided with OpenAI's Developers’ Day celebrations, signaling a significant step toward streamlining AI development for businesses and individual developers alike.

Before AgentKit, creating AI agents was a complex, time-consuming process. Developers had to juggle multiple tools, write custom code, and spend weeks designing functional workflows. AgentKit changes the game by providing a unified platform where developers can visually design workflows, test AI behaviour, and deploy chat-based agents without the typical headaches.

One of the standout features, Agent Builder, acts as a visual workflow tool. It enables developers to drag and drop components to create multi-agent workflows, allowing teams to visualise agent operations, test performance, and make iterative improvements quickly and efficiently.

To help organisations manage data seamlessly, OpenAI also introduced the Connector Registry. This feature integrates with popular apps and services such as Dropbox, Google Drive, Microsoft Teams, and SharePoint, streamlining data access across multiple workspaces. Additionally, AgentKit incorporates Guardrails, a security layer designed to prevent agents from performing harmful actions or leaking sensitive information.

For chat experiences, ChatKit allows developers to embed AI agents into websites and apps effortlessly. These chat-based agents can support a variety of functions, including customer service, onboarding, research, or acting as knowledge assistants.

Testing and evaluation have been significantly enhanced with the new Evals features. Developers can now automatically grade agents, assess entire workflows, and even test models from third-party providers. Large organisations like Carlyle and Box have already reported faster development cycles and improved accuracy thanks to these tools.

Another notable advancement is the expansion of reinforcement fine-tuning (RFT), which enables models to follow custom rules and make better decisions. RFT is already available for some models and is currently in beta for GPT‑5.

Regarding availability, ChatKit and Evals are ready for all developers, while Agent Builder and Connector Registry are currently in beta. All tools are included under standard API pricing, with OpenAI promising future enhancements, including a Workflows API and additional deployment options.

By combining these features, AgentKit aims to remove the barriers that previously slowed AI agent development, offering a comprehensive, secure, and user-friendly platform for developers and enterprises looking to innovate quickly in the AI space.