Promise Keepers

Uni at Your Assistance is a CopilotKit-powered retail self-service support experience that moves beyond traditional text chat by pairing a 30% conversational support panel with a 70% dynamic generative UI workspace. Instead of only answering customer questions in text, Uni uses agent output to render functional UI components such as recent order selection cards, delivery promise dashboards, package timelines, delivery proof sections, service recovery recommendations, and wrong-delivery claim forms.

The project is built with a Next.js / React frontend, FastAPI backend, LangGraph agent orchestration, OpenAI-powered intent detection and explanation generation, policy retrieval from markdown documents, CopilotKit, AG-UI, and A2UI-style declarative component payloads. The LangGraph agent uses tools to retrieve order data, carrier tracking data, customer profile context, and service policy guidance. It then determines the customer intent, evaluates delivery promise status, generates an empathetic customer-facing explanation, and selects the appropriate UI components from an approved A2UI component catalog.

The key innovation is that the agent does not simply chat with the customer. It drives the support workspace. For example, when a customer asks “Where is my order?”, the agent returns an orderSelection A2UI payload and the interface renders recent orders. When the customer asks why order U-1002 missed delivery, the agent streams AG-UI-style progress events, evaluates tracking milestones against the original promise date, retrieves service recovery policy, and renders a promise dashboard with delay reason, timeline, refund/coupon details, and delivery proof where available. If a customer reports that a delivered package was not received, the agent opens an interactive wrong-delivery claim form and confirms the claim with a 1–2 day investigation SLA.

Technically, the project includes working code for a FastAPI + LangGraph backend, a React A2UI renderer, an LLM UI planner, AG-UI-style streaming progress events, and an official AG-UI endpoint tested through CopilotKit’s HttpAgent path. The frontend uses a controlled component registry so the LLM can select from approved UI components without generating arbitrary React code. This makes the experience safer, more enterprise-ready, and more useful than a pure chatbot.

Uni demonstrates how generative UI can improve post-purchase retail support by combining conversation, live operational data, policy-aware reasoning, and interactive workflows into one unified customer experience.