With Agentkit, Openai finally makes easy the creation of AI agents

With Agentkit, Openai finally makes easy the creation of AI agents

Openai launches an SDK to develop AI agents, which is directly inspired by automation tools such as N8N or Zapier.

OPENAI launched this October 6 Agentkit, a digital suite to create, deploy and optimize artificial intelligence agents. Intended for developers, the SDK (Software Development Kit) aims to facilitate the design of complex automation by using a simple and functional graphical interface. Benchmarking capacity, pre-configured connectors, Guardrails… Agentkit arrives with all the components necessary for the creation of an agency workflow in production.

Three essential components: agent builder, Connector registry, chatkit

Agentkit arrives with three different components, designed for the creation of complex orchestration.

#1 Agent Builder, the main interface

Agent Builder is certainly the central component of the SDK. On the principle of N8N, Zapier, or even IFTTT, the tool offers a visual creation interface based on a system of connected nodes. Each node represents a workflow stage, whether it is a specialized agent, an external tool, or a control rule (If, Else, etc.). The construction of the workflow is done in Drag and Drop. The interface makes it possible to configure the inputs and outputs of each component, to test the execution in real time via a preview function, and to debug directly in the environment.

Agent Builder incorporates Guardrails modules to secure and supervise his agent, tools, MCP connectors … In short, all the components necessary to develop a complete workflow. Once the workflow is finalized, two options are available to developers: export it as a code to deploy it on their infrastructure, or integrate it directly via Chatkit.

#2 Connector Registry, a data control center

Connector Registry allows you to centralize data management in the agentic workflow. Administrators can manage pre-configured connectors such as Dropbox, Google Drive, SharePoint, Microsoft Teams for example or configure third-party MCP servers. The interest is to centralize data management through several workspaces, and in agent Builder. Connector registry is accessible on the web but it can also be required via an API.

#3 Chatkit, to deploy cat -based agents

“The deployment of cat user interfaces for agents can be surprisingly complex: continuous response management, management of discussion wires, display of model reasoning and design of attractive cat experiences”, recalls Openai. Chatkit is a module to integrate all of these constraints into a single simple and usual module. . The integration is done simply by passing the workflow identifier created in Agent Builder. It is thus possible to personalize the appearance of a cat, to manage the streaming of the answers, the threads, the display of the reasoning … The goal is to obtain a cat agent “production ready” quickly for any company.

Benchmarking and Fine-Tuning Reinforcement

Agentkit also arrives with Evals, a complete platform to quickly assess the capacities of your agent workflows. The latter makes it possible to quickly build evaluation datasets, to analyze your agent’s friction points, to automate the optimization of your prompts (based on human variables or notes) and finally to assess the capacities of models of other suppliers on a given workflow.

Agentkit finally has a module to fine-tuner the OPENAI models with reinforcement. Fine-tuning reinforcement (RFT) allows you to refine a model by rewarding it when it produces the expected results. The objective is to adapt the behavior of models to the specific needs of an agentic workflow. Available with O4-Mini and GPT-5, the RFT in particular makes it possible to train the model to call the right tools at the right time (Custom Tool Calls). It is also possible to define criteria that determine what a “good” answer to teach the model to optimize your behavior.

OPENAI announces that Agentkit is already available for all developers. Chatkit and Evals are in the final version while Agent Builder and Connector Registry remain in beta. No additional cost is to be expected: OPENAI invoices only API calls to models, according to the standard tariff grid.

Jake Thompson
Jake Thompson
Growing up in Seattle, I've always been intrigued by the ever-evolving digital landscape and its impacts on our world. With a background in computer science and business from MIT, I've spent the last decade working with tech companies and writing about technological advancements. I'm passionate about uncovering how innovation and digitalization are reshaping industries, and I feel privileged to share these insights through MeshedSociety.com.

Leave a Comment