Citigroup has debuted Citi Sky, an AI-powered wealth platform built on Google‘s full technology stack.  

The $1.7 trillion bank launched the tool at Google Cloud Next ’26 in Las Vegas on April 23, marking the first time a financial institution has deployed Google’s AI arm Google DeepMind‘s real-time avatar technology alongside Gemini’s Live API, JP Suh, product manager at Google DeepMind, told FinAi News. 

The tool enables low-latency audio and video conversations with an AI agent designed to work alongside human wealth advisers, not replace them, Joe Bonanno, global head of data analytics and innovation at Citi Wealth, told FinAi News. 

The platform  

Citi Sky is an AI platform allowing clients to talk to an AI avatar called ‘Sky’ that provides responses to client queries such as checking CD maturities, starting a college fund and researching whether refinancing a mortgage would be beneficial, according to Citi’s sample video. 

“Citi Sky operates as a true agentic platform,” Suh said. “The agent is capable of multistep reasoning, navigating meandering conversations, balancing shifting objectives in real-time and executing specific tasks — like parsing a mature certificate of deposit and seamlessly booking a calendar appointment with a specialist.” 

While other bank AI models are trained on general finance knowledge, the underlying model that powers Citi Sky is trained neither on any personal financial data nor on Citi’s proprietary enterprise data, Bonanno said. 

The goal of Citi Sky is to remove friction from day-to-day responsibilities for employees, so that they can spend more time with clients, he said. 

“During beta testing, we focused heavily on practical, adviser‑level impact,” Bonanno said. “Early results showed meaningful improvements in [employee] productivity, turnaround time on client-related tasks and the consistency of insights generated for advisers.” 

The tech 

Citi Sky runs on the Gemini Enterprise Agent Platform and is powered by advanced foundation models that have been built with strict evaluations and reliable tool use, rather than fine-tuned on proprietary client data, Suh said. 

“The development of Citi Sky involved Google’s full AI stack, from [Google] Cloud infrastructure to Google DeepMind,” Suh said. The tool was built with a “strategic partnership spanning over 12 months.” 

It is crucial to distinguish between how a model is trained and how an agent accesses data in real-time, Suh said. 

“The agent lives securely within Citi’s system, follows strict routing pathways and accesses only the appropriate amount of context at the exact moment it needs to answer a client’s question, he said.” 

Citi Sky’s agentic AI avatar “isn’t a one-off, but it was a first,” Rohit Bhat, managing director of financial services at Google Cloud, told FinAi News. “We are deepening our commitment to the financial sector with a singular goal: making Google Cloud the definitive AI stack for the industry.” 

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.