OpenAI Agent Builder & AgentKit – The Full Practical Guide

Build a weather app with agent builder by creating a step-by-step workflow, using classifiers, if-else logic, and widgets, with web search tools and JSON schemas.

Build interactive widgets to convert chat responses into a graphical card interface, including a cat, using widget studio and JSON data to render weather and other information.

Configure a weather widget in the agent builder, including hosting images and exporting output as JSON. Define required properties like background, images, temperature, and location, then debug schema mismatches.

Design a beautiful weather widget by selecting backgrounds and images using conditional logic or an agent, balancing performance costs, and refining image sizing for a polished card.

Publish and customize openai playground experiences with jet kit, adjusting accent color, fonts, greetings, and disclaimer, then embed a tailored chat widget on a website.

Embed OpenAI chat on your website with the recommended front-end Checkit integration, hosting the back end via Agent Builder, or run Checkit on your own infrastructure with the Python SDK.

Turn off chat kit outputs, sources, and annotations in OpenAI chat for a cleaner interface. Publish, refresh, and test the setup while reviewing the agent workflow and evaluation options.

Evaluate and optimize your agent performance by tracing bottlenecks, testing faster models like nano, and balancing reasoning and tool use to reduce time and cost.

Use graders to stress test chatbots by defining pass criteria, such as a weather widget output, and evaluate with scenarios and inputs to verify outcomes.

Connect an MCP server to Google Calendar within a new flow to find free slots and book calls. Configure access tokens, permissions, and guardrails to test and approve tool use.

Implement guardrails to protect user privacy in an AI agent workflow by detecting and removing personally identifiable information before it reaches the model, with configurable rules and tests.

Explore guardrails like moderation, jailbreak protection, and hallucination checks. These prevent unsafe content, block prompt injections, and validate outputs against a knowledge base via a vector store.

Learn how to augment AI agents with a vector store as a knowledge base, enabling semantic search over embeddings to pull company data and documents for tailored product recommendations.

Attach a vector store to your agent by uploading or selecting a knowledge base, then use semantic search to answer questions about shopping items and product data.

Master the use of variables and states to store and pass information across a multi-step workflow. Save and reuse values like prices and inputs to drive outputs and user approvals.

Master managing states in shopping assistants by implementing a user approval workflow and maintaining current item name, price, and total price in state for cart updates.

Create an AI growth assistant that handles cold and ready leads, delivers a lead magnet, and schedules calls via Zapier and Google Calendar, using an intent classifier and time-slot workflow.

Set up a Zapier integration with the MCP server to automate scheduling calls in Google Calendar, deliver lead magnets to cold leads, and route warm leads based on user intent.

Build a classifier agent to identify user intent for lead magnet or schedule call, design a JSON output, and implement stateful routing with conditional logic.

Upload the lead magnet PDF to Google Drive, make the link public or downloadable, paste it into the widget prompt, then build, test, and plan a call.

OpenAI agent builder & agentkit: schedule a Google Calendar integration to suggest 30-minute time slots over five business days and create calendar events via native or Zapier MCP tool.

Polish the agent workflow by implementing a time slot selection state, json outputs with schema, and robust scheduling logic, while troubleshooting optional fields and ending the workflow to prevent loops.