OpenAI has launched AgentKit, a comprehensive stack designed for building, deploying, and tuning AI agents. This kit integrates several components: a visual Agent Builder, an enterprise Connector Registry, an embeddable UI toolkit named ChatKit, expanded Evals, and reinforcement fine-tuning options. The primary users targeted by this offering are developers and enterprise teams who require governed and measurable agent workflows across various products and organizations.
BREAKING 🚨: OpenAI announced Agent Builder during DevDay, and it is now already available on OpenAI Platform!
— TestingCatalog News 🗞 (@testingcatalog) October 6, 2025
Testing time! 🔥 https://t.co/fFf4NWC2Nq pic.twitter.com/qcMpE9G1e2
The Agent Builder features a drag-and-drop canvas equipped with versioning, preview runs, and inline guardrails. It includes templates that cover common workflows. The Connector Registry provides administrators with a single panel to manage data sources across ChatGPT and the API, featuring prebuilt connectors and third-party MCPs. ChatKit allows for embedding a production chat experience with threading and streaming capabilities, enabling teams to integrate agent UIs within applications and websites. Evals has been expanded to include datasets, trace grading, a prompt optimizer, and support for testing third-party models. Reinforcement fine-tuning is generally available on o4-mini and in private beta on GPT-5, offering custom tool-call training and custom graders.
ChatKit Studio from OpenAI is now available as well, with Playground, Widget builder and World demo app.
— TestingCatalog News 🗞 (@testingcatalog) October 6, 2025
Widget gallery is featuring @patience_cave 🔥 pic.twitter.com/E4BLe5fkQs
Availability begins today: ChatKit and the new Evals are generally available. Agent Builder is currently in beta. The Connector Registry is entering beta for customers utilizing the new Global Admin Console. Pricing is aligned with standard API model rates. Early adopters have reported faster build times and improved accuracy in production agents.
AgentKit builds upon OpenAI’s Responses API and Agents SDK, which were introduced earlier this year. The company presents it as an end-to-end solution: design with Agent Builder, govern through the Registry, deploy using ChatKit, and then measure and optimize with Evals and fine-tuning. OpenAI has indicated that more features are forthcoming, including a Workflows API and agent deployment options within ChatGPT.
