Google tests Remy, a staff‑only Gemini agent that could act on users' behalf

Google is internally trialing Remy, a proposed 24/7 personal agent for Gemini that would move the product from chat responses toward acting on users’ behalf.

Google is testing an AI personal agent called Remy inside a staff‑only version of the Gemini app, according to reporting based on an internal document and conversations with two people familiar with the project. The trial is limited to employees, the material says, and Google declined to comment. The report does not provide a public release date or timetable beyond the internal trial.

Internal materials describe Remy as a “24/7 personal agent” designed to shift Gemini from a chat interface into an assistant that can act on a user’s behalf. The writeup frames Remy as able to monitor and manage complex tasks, learn and apply user preferences, and integrate with the services most relevant to a user rather than only returning conversational replies.

Remy is described as a further step beyond Google’s existing agent features — such as Agent Mode, whose availability varies by subscription tier and region. The concept was compared to OpenClaw, an autonomous agent that drew attention earlier this year for sending replies and performing research tasks; reporting has said OpenAI hired OpenClaw’s creator. Google has not confirmed whether Remy would match that level of autonomy.

Google’s Gemini support and privacy documentation outline the connected app surface and user controls that would matter if an agent like Remy were expanded. Connected Apps include Google Workspace services — Gmail, Calendar, Docs, Drive, Keep and Tasks — and third‑party integrations listed in help pages such as GitHub, Spotify, YouTube Music, Google Photos, WhatsApp, Google Home and Android utilities. it does not disclose Remy’s underlying model version or architecture, whether Remy can act without explicit user confirmation, or how actions and approvals would be logged. The internal writeup frames this as a dog‑fooding project, with employees testing before any broader release.

The trial underscores control, privacy and governance questions builders and users should watch. Gemini’s Privacy Hub currently lets users review and delete Gemini Apps Activity, set auto‑delete preferences, choose whether their data is used to improve Google AI, and control saved personalization data. Guidance cited from Google Research and Google Cloud emphasizes well‑defined human controllers, least‑privilege permissions, observable actions and transparent, auditable logging for agent activities — measures that would be especially relevant if an always‑on agent is rolled out more widely.