October 2025 Summaries

AG-UI Protocol, an open protocol facilitating real-time collaboration between humans and agents across various applications, platforms, and frameworks, has become a pivotal part of the evolving agentic ecosystem. Initially developed as an internal solution within CopilotKit, it has expanded into a widespread open-source initiative, emphasized by its rapid growth in adoption and integration by large-scale industry teams. The Protocol Triangle, comprising MCP, A2A, and AG-UI, forms the backbone of this ecosystem by enabling communication between agents, tools, and users, fostering a collaborative environment where these entities can interact and learn together in real time. Recent milestones include Google’s ADK joining the AG-UI ecosystem, a significant increase in agent-user interactions, and remarkable growth in GitHub stars and weekly installations, indicating the protocol's swift acceptance and potential for standardizing the agentic web. As the CopilotKit team continues to expand SDK support and enhance integration capabilities, AG-UI stands out as a crucial protocol for developers building collaborative systems involving humans and agents.

The text provides a detailed guide on building a full-stack AI shopping assistant using CrewAI Flows and the Tavily Search API for the backend, paired with CopilotKit for the frontend. The AI assistant is designed to search for products on platforms like Amazon, Target, and eBay, extracting structured product data and offering tailored recommendations. CrewAI Flows facilitate the automation of tasks by linking AI agents and managing data sharing, while the Tavily API enhances search capabilities. The setup requires knowledge of React or Next.js, Python, and access to GPT-4 via the OpenAI API. The frontend is developed using CopilotKit, which aids in state management, human-in-the-loop functionality, and UI updates. The guide emphasizes creating a seamless integration between the backend and frontend, detailing steps for configuring the CopilotKit runtime, setting up chat components, and synchronizing agent states. The tutorial highlights the potential of CopilotKit in developing interactive AI applications and encourages further exploration of its capabilities.

A CopilotKit tutorial explains how to build a full-stack travel-planning multi-agent application that combines Google’s Agent-to-Agent (A2A) Protocol, CopilotKit’s AG-UI Protocol, Google ADK, and LangGraph. A central orchestrator agent built with Google ADK uses Gemini to coordinate specialized itinerary, restaurant, weather, and budget agents, including agents implemented with different frameworks and models, while A2A agent cards, skills, executors, and HTTP servers make those services discoverable and callable. The guide covers creating the project with the CopilotKit CLI, configuring Python and frontend dependencies and API keys, exposing the orchestrator through FastAPI and AG-UI, and running each agent as an A2A service. On the Next.js frontend, CopilotKit middleware connects the chat interface to the orchestrator and remote agents, displays real-time inter-agent requests and responses through generative UI, extracts structured results into itinerary, weather, restaurant, and budget components, and adds a human-in-the-loop budget-approval workflow that pauses execution until a user accepts or rejects a proposal.

The text serves as a comprehensive guide on building fullstack AI agents using Pydantic AI and the AG-UI protocol, with integration into CopilotKit for frontend interaction. The AG-UI protocol, developed by CopilotKit, facilitates real-time, event-driven communication between the frontend and AI agents, making it possible to manage state, stream responses, and utilize tools effectively. The guide walks through the process of integrating Pydantic AI agents with the AG-UI protocol, setting up a Python-based backend and a Next.js frontend, and utilizing various tools and events to maintain state and handle AI agent tasks. It further explains how to create a frontend using CopilotKit, set up chat components, synchronize agent state with the frontend, and implement Human-in-the-Loop functionality for enhanced AI-human collaboration. Additionally, it provides insights on streaming AI agent responses in the frontend, illustrating the capabilities of CopilotKit in building interactive AI applications. The document concludes by highlighting the potential applications of CopilotKit in developing agentic solutions and encourages readers to explore its features further.