OpenAI just launched AgentKit, a full toolkit for developers and enterprises to build, deploy, and improve AI agents.
There are some similar tools out there, most known lately surely n8n.io, but if OpenAI is set to launch this tool, this means a lot more people will have access and will try “agentic” technology and not just generative one.
The toolkit includes three main parts:
Agent Builder lets you design and manage multi-agent workflows visually. You’ll be able to connect tools, guardrails, and version your setup—all on a drag-and-drop canvas. Teams like Ramp and LY Corporation built complete agents in just a few hours, cutting months of orchestration work.
Connector Registry gives enterprises one admin panel to manage all their data connections—Dropbox, Google Drive, SharePoint, Teams, and more—across ChatGPT and the API. It supports guardrails to protect data and detect risky behavior.
ChatKit simplifies embedding chat-based agents directly into apps and websites. It handles streaming responses, conversation threads, and UI customization out of the box. HubSpot used it to launch Breeze Assistant and reduce weeks of frontend work.
OpenAI also expanded its Evals platform with tools for testing and improving agent performance.
Developers will now be able to use datasets, trace grading, automated prompt optimization, and even evaluate third-party models. Companies like Carlyle and Box report major gains in accuracy and speed.
Finally, Reinforcement Fine-Tuning (RFT) is now generally available on o4-mini and in beta for GPT-5, allowing developers to train models with custom tools and grading logic for better reasoning and precision.
ChatKit and Evals are available now. Agent Builder and Connector Registry are in beta. All come with standard API pricing.
AgentKit is OpenAI’s clearest move toward making agent creation as simple and reliable as software development itself.
See the full article.