Summary
Riley Brown walks through building a fully functional voice-controlled AI desktop companion — nicknamed “Ricky” — using Cursor, OpenAI’s GPT Realtime 2, and the Exa web search API, with no prior coding experience required. The tutorial demonstrates how a series of natural-language prompts to Cursor’s agent window can generate, debug, and iteratively refine a working application from an empty folder in a single session.
The finished agent features an animated face with expressive mood states and lip-synced speech, a split-screen artifact panel for displaying structured outputs like Mermaid charts and search results, and a full tool suite including Exa-powered web search, image generation, and computer use — enabling Ricky to open applications, control the browser, and interact with the desktop autonomously on voice command. GPT Realtime 2 drives the conversation layer, supporting low-latency interruption and natural back-and-forth dialogue. Brown covers the API key setup for both OpenAI and Exa, and shares the full prompts in the video description.
Several build iterations are shown in real time: improving the markdown rendering of web search results, tuning facial animation quality, adding a menu command to list available capabilities, and adjusting the panel layout. The video makes a practical case that vibe coding with modern AI IDEs has compressed what would once have been a multi-day agentic application project into an accessible afternoon build.
📺 Source: Riley Brown · Published July 01, 2026
🏷️ Format: Hands On Build
