OpenAI Unveils AgentKit and Visual Agent Builder to Streamline AI Agent Production
A visual-first stack for agentic workflows
OpenAI announced AgentKit, a unified platform that brings a visual Agent Builder, an embeddable ChatKit UI, an Agents SDK, and expanded Evals into a single workflow aimed at shipping production AI agents faster. Agent Builder is rolling out in ‘beta’ while other components are generally available.
What Agent Builder offers
Agent Builder provides a drag-and-drop canvas for composing multi-step, multi-agent workflows. Users can connect nodes, attach per-node guardrails, preview runs, configure inline evaluations, and maintain full versioning. Teams can begin from templates or a blank canvas, with execution handled by the Responses API. The visual approach aims to compress iteration cycles when moving from prototype to production.
Agents SDK: code-first alternative
For teams that prefer code, the Agents SDK offers type-safe libraries in Node, Python, and Go. It is positioned as a faster integration path than manually orchestrating prompts and tools, while sharing the same execution substrate as the canvas via the Responses API.
ChatKit and built-in connectors
ChatKit, now GA, is a drop-in, brand-customizable chat interface for web and app deployments. It supports streaming, threaded conversations, and thinking UIs, helping teams deliver a production chat surface without building a frontend from scratch. Agent workflows can call built-in tools such as web search, file search, image generation, a code interpreter, and a feature OpenAI calls computer use. External connectors, including Model Context Protocol servers, reduce custom glue code for common tasks.
Connector Registry and governance
OpenAI introduces a Connector Registry in ‘beta’ to centralize admin governance across ChatGPT and the API. The registry manages data sources like Dropbox, Google Drive, SharePoint, Microsoft Teams, and third-party MCPs. Rollout begins for customers with access to the Global Admin Console, enabling admins to govern connections and data flows.
Evals and continuous optimization
The Evals suite is generally available with new capabilities: dataset support, trace grading for end-to-end workflow assessment, automated prompt optimization, and third-party model evaluation. These features let teams instrument agents with continuous measurement and iterate on prompts and designs based on graded traces.
Pricing and availability
ChatKit and the new Evals features are GA, while Agent Builder is available in ‘beta’. All components use the standard API model pricing, meaning customers pay for model and compute usage rather than separate product SKUs.
How the components fit together
Design: Construct agents visually in Agent Builder or programmatically with the Agents SDK, both executing over the Responses API. Deploy: Embed agentic experiences with ChatKit to provide a ready-made front end. Optimize: Use Evals for datasets, trace grading, and automated prompt tuning to increase task accuracy over time.
Safety and operational controls
Agent Builder pairs with open-source, modular guardrails that can detect jailbreak attempts, mask or flag PII, and enforce policies at node and tool boundaries. The Connector Registry provides centralized governance for data sources, spanning both ChatGPT and API usage.
Practical assessment
AgentKit packages the core pieces teams need to move agentic prototypes into production: versioned node graphs, built-in tools and connectors, centralized governance, and standardized evaluation hooks. The primary value is operational, reducing bespoke orchestration and frontend work while keeping evaluation in the development loop.