OpenAI introduces new toolkit to help you build AI agents
What's the story
OpenAI has launched AgentKit, a new toolkit for creating and deploying artificial intelligence (AI) agents. The announcement was made by CEO Sam Altman at the company's Dev Day event. "AgentKit is a complete set of building blocks available in the OpenAI platform designed to help you take agents from prototype to production," Altman said.
Toolkit features
Key components of AgentKit
AgentKit is designed to simplify the process of building, deploying, and optimizing agent workflows. It comes with two main components: Agent Builder and ChatKit. The former lets developers create agents visually by outlining their logic and steps quickly. The latter offers a simple chat interface that can be integrated into apps, allowing developers to customize it according to their brand's unique workflows.
Performance evaluation
Evals for agents
AgentKit also includes Evals for Agents, a suite of tools to evaluate the performance of AI agents. This includes step-by-step trace grading, datasets for evaluating individual components of an agent, automated prompt optimization, and the ability to evaluate external models from the OpenAI platform. These features are designed to give developers a comprehensive view of their agents' performance and areas for improvement.
Security features
Connector registry
AgentKit also provides access to OpenAI's connector registry. This lets developers connect agents to internal tools and third-party systems through an "admin control panel," all while keeping security and control intact. This feature is particularly useful for those looking to integrate their AI agents with other platforms or services without compromising data security.
Live demonstration
Altman says AgentKit will accelerate agent development
To demonstrate the simplicity of AgentKit, OpenAI engineer Christina Huang built an entire AI workflow and two agents onstage in under eight minutes. "This is all the stuff that we wished we had when we were trying to build our first agents," Altman said. He also revealed that several launch partners have already scaled agents using AgentKit, further proving its effectiveness in real-world applications.