Google tests Remy AI agent internally, designed to act autonomously across Gemini services

Google is testing Remy, an AI personal agent for Gemini that can take actions on users' behalf across Google services, according to a Business Insider report citing an internal document and two sources familiar with the project. The tool is currently in employee-only testing with no confirmed public release date.

What Remy does

According to Google's internal description, Remy is positioned as a "24/7 personal agent" designed to transform Gemini from a chat assistant into one that can act autonomously. The system is described as more advanced than Google's existing Agent Mode and is designed to integrate with Google services, monitor relevant information for users, and learn preferences over time.

The report did not provide technical details on Remy's architecture, the specific Gemini model version powering it, or the level of autonomy being tested. Key unanswered questions include whether Remy can act independently without user confirmation and how it handles action approvals and logging.

Connected services scope

Gemini's current connected services include Google Workspace (Gmail, Calendar, Docs, Drive, Keep, Tasks), plus GitHub, Spotify, YouTube Music, Google Photos, WhatsApp, Google Home, and Android utilities, according to Google's support documentation. The report did not specify which services are included in the current Remy employee test.

Control and privacy framework

Google's Gemini Privacy Hub allows users to review and delete Gemini Apps Activity, adjust auto-delete settings, and manage whether data improves Google AI models. Users can also control access to connected apps and saved information.

Google Research has stated that AI agents should have well-defined human controllers, carefully limited powers, observable actions, and planning ability. Google Cloud emphasizes transparent and auditable agent activities through logging, with powers limited according to intended purpose and user risk tolerance.

Internal testing approach

The internal document characterizes Remy as a "dog-fooding project," standard industry terminology for employee testing before broader release. A Google spokesperson declined to comment on the report.

The report compared Remy's concept to OpenClaw, an AI agent that gained attention earlier this year for autonomously replying to messages, conducting research, and taking actions. OpenAI CEO Sam Altman said in February that OpenAI was hiring OpenClaw's creator, according to the report.

What this means

Remy represents Google's push beyond conversational AI into agentic systems that can execute tasks autonomously. The lack of technical details—particularly around autonomy levels, approval mechanisms, and action logging—indicates this is early-stage internal testing. The emphasis on user control and privacy frameworks suggests Google is attempting to address the trust and safety concerns that have plagued other autonomous agent projects. Without confirmed release plans or technical specifications, Remy's actual capabilities and limitations remain unclear.