OpenAI Debuts Agent Builder and AgentKit: A Visual-First Stack for Building, Deploying, and Evaluating AI Agents

OpenAI Debuts Agent Builder and AgentKit: A Visual-First Stack for Building, Deploying, and Evaluating AI Agents

Estimated reading time: 6 minutes

• OpenAI has launched AgentKit, a comprehensive, visual-first platform simplifying the entire lifecycle of AI agent development, from building to evaluation.
• AgentKit integrates Agent Builder (visual composer), Agents SDK (code-first), ChatKit (embeddable UI), built-in tools, a Connector Registry, and enhanced Evals.
• The platform promotes a synergistic workflow: Design with Agent Builder or SDK, Deploy seamlessly with ChatKit, and Optimize continuously using Evals.
• It incorporates robust safety and governance features through guardrails and the Connector Registry, ensuring secure and compliant AI agent operations.
• AgentKit is priced under standard API model pricing and has already shown real-world impact, enabling businesses like Albertsons to leverage context-aware AI for operational insights.

• Unveiling AgentKit: A Unified Ecosystem for AI Agent Development
  • Agent Builder (beta): The Visual Composer
  • Agents SDK: Code-First Flexibility
  • ChatKit (GA): Seamless Deployment UI
  • Built-in Tools and Connectors: Expanding Agent Capabilities
  • Connector Registry (beta): Centralized Governance
  • Evals (GA) and Optimization: The Continuous Improvement Loop
  • Pricing and Availability
• The Synergistic Workflow: Design, Deploy, Optimize
• Safety and Governance: Baked into AgentKit’s Core
• Real-World Application: Albertsons’ Ice Cream Sales Agent
• Three Actionable Steps for Developers and Businesses
• Conclusion
• Start Building Your Intelligent Agents Today!

The landscape of artificial intelligence is evolving at a breakneck pace, with AI agents moving from theoretical concepts to practical, impactful applications. Yet, the journey from ideation to a production-ready agent has often been fraught with complexity, requiring extensive coding, custom integrations, and fragmented evaluation processes. Recognizing this challenge, OpenAI has unveiled a groundbreaking solution designed to streamline the entire lifecycle of AI agent development.

Enter AgentKit, a powerful new offering poised to redefine how developers and businesses build, deploy, and optimize intelligent agents. By consolidating essential tools into a cohesive, visual-first stack, OpenAI aims to democratize agent creation, making advanced AI capabilities more accessible and efficient than ever before.

This comprehensive platform promises to accelerate innovation, allowing teams to focus on the strategic value of their agents rather than getting bogged down in intricate infrastructure. Let’s delve into what AgentKit brings to the table and how it’s set to transform the future of AI.

Unveiling AgentKit: A Unified Ecosystem for AI Agent Development

The core of OpenAI’s latest innovation lies in its holistic approach. OpenAI has released AgentKit, a cohesive platform that packages a visual Agent Builder, an embeddable ChatKit UI, and expanded Evals into a single workflow for shipping production agents. The launch includes Agent Builder in beta and the rest generally available. This unified vision brings together several critical components, each meticulously designed to address specific pain points in the agent development journey.

Agent Builder (beta): The Visual Composer

At the heart of AgentKit’s visual-first philosophy is the Agent Builder. This intuitive, drag-and-drop canvas empowers developers to compose multi-step, multi-agent workflows with unprecedented ease. Teams can start from pre-built templates or a blank canvas, connecting tools, defining logic, and implementing per-node guardrails through a graphical interface. The ability to conduct preview runs, configure inline evaluations, and utilize full versioning means faster iteration cycles and a smoother transition from prototype to production. This visual approach significantly compresses the time and effort traditionally required for complex agent orchestration.

“With Agent Builder, you can drag and drop nodes, connect tools, and publish your agentic workflows with ChatKit and the Agents SDK,” tweeted OpenAI Developers, highlighting the platform’s user-friendly nature. This visual workflow is powered by the Responses API, ensuring robust execution.

Agents SDK: Code-First Flexibility

For developers who prefer a code-centric approach or need deeper programmatic control, AgentKit offers the Agents SDK. This code-first alternative provides type-safe libraries in popular languages like Node, Python, and Go. OpenAI positions the SDK as a faster way to integrate agentic capabilities compared to manual prompt-and-tool orchestration, all while leveraging the same powerful execution substrate as the Agent Builder – the Responses API. This dual approach ensures that AgentKit caters to a wide spectrum of development preferences.

ChatKit (GA): Seamless Deployment UI

Building a robust, user-friendly frontend for an AI agent can be a significant undertaking. ChatKit, now generally available, solves this by offering a drop-in, brand-customizable chat interface. This embeddable UI is designed for deploying agentic experiences directly into web applications or mobile apps. It inherently handles complex functionalities like streaming responses, managing conversation threads, and providing engaging “thinking” UIs, eliminating the need for developers to build these features from scratch. OpenAI’s marketing materials showcase organizations successfully using ChatKit for everything from customer support to internal assistants.

Built-in Tools and Connectors: Expanding Agent Capabilities

Agent workflows often require interaction with various external services and data sources. AgentKit addresses this by including a suite of built-in tools and connectors. Agents can effortlessly call upon web search, file search, image generation, code interpreter, and even “computer use” functionalities. Furthermore, support for external connectors, including Model Context Protocol (MCP) servers, significantly reduces the ‘glue code’ traditionally needed for common tasks, enabling agents to be more capable and autonomous right out of the box.

Connector Registry (beta): Centralized Governance

Managing data sources and ensuring secure, compliant access is paramount for enterprise AI deployments. The Connector Registry, currently in beta, provides centralized administrative governance across both ChatGPT and the API. This feature allows organizations to manage connections to critical data sources such as Dropbox, Google Drive, SharePoint, and Microsoft Teams, as well as third-party MCPs. Its rollout, beginning for customers with the Global Admin Console, ensures that data flows and connections adhere to organizational policies and security standards.

Evals (GA) and Optimization: The Continuous Improvement Loop

For any AI agent to be truly effective, continuous measurement and optimization are essential. AgentKit’s Evals, now generally available, introduces new capabilities designed to raise task accuracy and refine agent performance. These include comprehensive datasets, trace grading for end-to-end workflow assessment, automated prompt optimization, and even third-party model evaluation. OpenAI emphasizes that these tools facilitate a continuous measurement process, allowing teams to identify weaknesses, iterate on prompts, and incrementally improve agent reliability and effectiveness.

Pricing and Availability

OpenAI has confirmed that ChatKit and the new Evals features are generally available, while Agent Builder is currently in beta. Importantly, all these new components are included under standard API model pricing, meaning users will pay for model and compute usage rather than separate SKUs. This transparent pricing model simplifies cost management for developers and businesses.

The Synergistic Workflow: Design, Deploy, Optimize

AgentKit’s true power lies in how its individual components interlock to form a coherent, efficient workflow:

• Design: Developers can leverage the visual Agent Builder to graphically assemble agents and define guardrails, or opt for the Agents SDK to programmatically write agents against the Responses API. This flexibility caters to diverse skill sets and project requirements.
• Deploy: Once an agent is designed, ChatKit provides a seamless path to production. By embedding this customizable chat UI, organizations can deliver intelligent agent experiences to their users without the overhead of building a frontend from scratch.
• Optimize: The journey doesn’t end at deployment. AgentKit integrates Evals with robust capabilities like datasets, trace grading, and automated graders. This allows teams to continuously instrument their agents, gather performance data, and iterate on prompts based on graded traces, ensuring ongoing improvement in task accuracy.

This consolidated stack significantly reduces the need for bespoke orchestration and custom frontend development. OpenAI’s assessment highlights that the value is profoundly operational: versioned node graphs, built-in tools (web/file search, computer use), centralized connector governance, and standardized evaluation hooks address critical production concerns that previously demanded custom infrastructure and considerable engineering effort.

Safety and Governance: Baked into AgentKit’s Core

Recognizing the paramount importance of responsible AI, OpenAI has integrated robust safety and governance features directly into AgentKit. The launch materials emphasize that Agent Builder works hand-in-hand with guardrails – open-source, modular components designed to detect potential issues like jailbreaks, mask or flag Personally Identifiable Information (PII), and enforce predefined policies at the individual node and tool boundaries. Furthermore, the Connector Registry plays a crucial role in empowering administrators to govern connections and control data flows across both ChatGPT and the API, ensuring a secure and compliant environment for all agentic applications.

Real-World Application: Albertsons’ Ice Cream Sales Agent

The practical implications of AgentKit are already taking shape. A compelling example comes from Albertsons, one of the largest food and drug retailers in the United States. They utilized AgentKit to build an intelligent agent designed to assist associates with sales strategy. Imagine an associate asking the agent to create a plan to improve ice cream sales. The agent, powered by AgentKit, immediately processes the full context – accounting for seasonality, historical sales trends, and external market factors – to provide a tailored, data-driven recommendation. This showcases AgentKit’s ability to facilitate context-aware, actionable intelligence directly within business operations, empowering employees with powerful insights on demand.

Three Actionable Steps for Developers and Businesses

Ready to harness the power of AgentKit? Here’s how you can get started:

1. Explore the Visual Builder or Agents SDK: Begin by experimenting with the Agent Builder for rapid, visual prototyping of your multi-step workflows. If your project demands deeper programmatic control, dive into the Agents SDK with its type-safe libraries in Node, Python, or Go. Start small, understand the core concepts, and gradually expand your agent’s capabilities.
2. Leverage ChatKit for Rapid Deployment: Once your agent logic is sound, don’t waste time building a custom chat interface from scratch. Utilize ChatKit’s embeddable, customizable UI to quickly deploy your agentic experiences on your website or within your applications. This accelerates user adoption and feedback collection significantly.
3. Implement Continuous Evaluation with Evals: From day one, integrate AgentKit’s Evals capabilities into your development cycle. Define datasets, set up trace grading, and leverage automated prompt optimization. Continuous measurement is key to iteratively improving your agent’s accuracy, reliability, and overall performance in real-world scenarios.

Conclusion

OpenAI’s AgentKit represents a pivotal moment in the evolution of AI. By offering a consolidated, visual-first stack that encompasses building, deploying, and evaluating AI agents, OpenAI is significantly simplifying what was once a complex, resource-intensive process. From the intuitive Agent Builder and robust Agents SDK to the seamless deployment with ChatKit and the continuous optimization offered by Evals, AgentKit provides a comprehensive toolkit for bringing sophisticated AI agents to life.

This platform promises to democratize AI agent development, enabling businesses and developers to create more intelligent, capable, and trustworthy autonomous systems with unprecedented speed and efficiency. The operational value, coupled with integrated safety and governance, positions AgentKit as a foundational tool for the next wave of AI innovation.

Start Building Your Intelligent Agents Today!

The future of AI agents is here, and it’s more accessible than ever. Visit the OpenAI Developers website to explore AgentKit, sign up for Agent Builder beta access, and dive into the documentation. Begin your journey toward building powerful, production-ready AI agents that can transform your operations and user experiences.

Frequently Asked Questions (FAQ)