Google has started internal tests of a new AI agent named Remy inside a private version of its Gemini app. The project aims to create a constant companion that handles work tasks, school assignments, and daily routines without constant prompts. An internal description calls it a 24/7 personal agent powered by Gemini that takes actions on a user’s behalf, watches for important updates, manages complex workflows proactively, and adapts to individual habits over time. Business Insider first reported the details after reviewing company documents and speaking with people close to the effort. Google declined to comment on the project.
Development comes at a telling moment. OpenClaw, the open-source agent released earlier this year, quickly gained more than 100,000 GitHub stars and drew praise from Nvidia chief executive Jensen Huang, who called it definitely the next ChatGPT. Demand for the tool even lifted secondhand MacBook prices 15 percent in China, according to reports. OpenAI later hired its creator. Those events showed clear appetite for agents that operate across a computer with minimal oversight.
Google’s version stays within its own services for now. Remy connects to Gmail, Docs, Calendar, Drive, and Search so it can book meetings, draft replies, organize files, and follow up on research threads. Employees in the dogfooding phase use it daily to surface how it learns preferences and reduces manual steps. The company ended its earlier Mariner agent experiment on May 4 and folded that team’s work into the new effort, The Decoder noted.
The competitive field has filled quickly
Anthropic already offers Claude Cowork for desktop tasks. Meta acquired Manus AI and released My Computer, which sorts files, launches applications, and sends messages on command. Nvidia introduced NemoClaw as an open platform for businesses to run agents on varied hardware. Security researchers have flagged risks in OpenClaw, including exposed admin panels and prompt-injection weaknesses. Google’s approach, built inside a trusted consumer app, could sidestep some of those concerns through tighter integration and existing privacy controls.
Industry watchers expect agents to feature at Google I/O on May 19 and 20. The event often sets the tone for the company’s AI direction. Remy may appear in demo form if the internal build reaches a stable point. Recent coverage from PYMNTS and LiveMint confirms the same timeline and description, while X discussions show developers debating whether users will trust an agent with full inbox and calendar access.
Google’s move signals a broader shift from chat-based tools to systems that act independently. The company has watched rivals ship working agents and now positions Remy as its polished, always-available answer. No public release date exists yet, but the internal momentum suggests the project will move beyond employee testing soon.
