Agent Builder in the OpenAI Platform

A canvas for multi-step agents

Agent Builder is OpenAI's visual canvas for building agent workflows: you drag nodes (agents, tools, logic, data steps) onto a canvas, connect them, preview runs with traces, and publish a versioned workflow. It was introduced as part of OpenAI's AgentKit tooling alongside the Agents SDK (code library for agents), ChatKit (embeddable chat UI for workflows), a connector registry for tools and data, and evals for testing agents. A workflow can be called from your app by its ID or exported as Agents SDK code you host yourself. Product details such as node names and menus change; check the current OpenAI documentation.

Canvas, code and deployment

A visual builder lets you design multi-step agent workflows, test them and publish versions that apps can call.

A flowchart that actually runs

Teams have always drawn agent flows on whiteboards. A workflow builder makes the drawing executable: each box is a real step and each arrow a real transition.

Treat the canvas as a design and test tool

Even if production runs in code, sketching and previewing the flow on a canvas is a fast way to agree on behaviour with non-developers.

Quick check: What is Agent Builder?

• A vector database
• A model training service
• A visual canvas for designing, testing and publishing agent workflows
• A replacement for HTTP