Google is reportedly testing an advanced new AI personal agent for Gemini, codenamed "Remy," which is designed to empower the AI to take proactive actions for users in various work and daily tasks. According to an internal document reviewed by Business Insider and information from two individuals familiar with the project, Remy is currently undergoing a "dog-fooding" phase, meaning it is being tested internally by Google employees within a staff-only version of the Gemini app.

Described internally as a "24/7 personal agent," Remy's core objective is to transform Gemini into a more capable assistant that can operate on a user's behalf. This initiative signifies Google's broader strategy to evolve Gemini beyond simple chat-based responses, extending its functionality into active task completion. While Google already offers agent-related features, such as "Agent Mode" with varying access based on subscription and region, Remy is depicted as a more sophisticated evolution. It is designed for deeper integration with Google services, monitoring user-relevant information, handling complex tasks, and learning individual user preferences to provide a highly personalized experience.

The current scope of Gemini's connected services, as detailed in Google's support documentation, includes a wide array of applications that allow it to fulfill user requests and deliver pertinent responses. These "Connected Apps" encompass Google Workspace services like Gmail, Calendar, Docs, Drive, Keep, and Tasks, along with third-party integrations such as GitHub, Spotify, YouTube Music, Google Photos, WhatsApp, Google Home, and various Android utilities. Remy is expected to leverage and expand upon this interconnected ecosystem.

Privacy and user control are central to the development of such AI agents, and Google's Gemini Privacy Hub addresses these concerns. It is designed to provide users with context and control over how Gemini interacts with connected apps, both Google's own and third-party services. Users will have the ability to review and delete their Gemini Apps Activity, modify auto-delete settings, and manage whether their data contributes to improving Google AI. Furthermore, the Privacy Hub will enable users to manage access permissions for other applications and data, as well as control information they have specifically requested Gemini to save.

Google Research and Google Cloud have established guidelines for AI agents, emphasizing the importance of well-defined human controllers, carefully limited powers, observable actions, and robust planning capabilities. These guidelines also advocate for transparency and auditability of agent activities through logging and clear action characterization, stressing the "least-privilege principle" to limit an agent's powers based on its intended purpose and user risk tolerance. Remy's reported preference-learning function underscores the criticality of memory controls, which the Gemini Privacy Hub addresses through options for managing saved information, personalization based on past chats, and "Personal Intelligence."

Despite the insights from the internal report, several key details about Remy remain undisclosed. The report did not specify a public release date or even confirm if Google intends to make Remy publicly available. It also refrained from identifying which particular Google services are part of the current employee testing phase. Crucially, technical specifics such as Remy's architectural design, the underlying model version, or the precise level of autonomy being tested were not provided. The report also did not clarify whether Remy can act independently without requiring explicit user confirmation for every action. These unanswered questions mean the mechanisms for approvals and logging of completed actions in Remy's current iteration are unclear.

The concept behind Remy has drawn comparisons to "OpenClaw," another AI agent that gained attention for its capabilities in autonomously replying to messages, conducting research, and performing actions on behalf of users. Notably, OpenAI CEO Sam Altman reportedly hired OpenClaw's creator in February. While Google DeepMind CEO Demis Hassabis has publicly discussed the vision of developing a powerful digital assistant, Google has yet to officially confirm whether Remy will transition into a public feature within Gemini.