Google Tests Remy AI Agent for Gemini App
Google started testing Remy, an AI personal agent built for its Gemini platform. Business Insider reported this development based on an internal document and interviews with two individuals close to the project. Remy aims to manage actions for people during work duties and everyday activities.
Internal Testing Underway
Employees at Google are using Remy in a version of the Gemini app limited to staff. The document calls Remy a "24/7 personal agent." It positions the tool to make Gemini function as an assistant that performs tasks for users. A Google representative chose not to comment on the matter. No timeline exists for a public launch, and the report skips details on specific Google services involved in the employee trials.
Remy fits into Google's efforts to grow Gemini past simple chat replies. The company already provides features like Agent Mode. Access to these depends on subscription levels and locations. Sources describe Remy as a step forward. It connects deeply with Google services, tracks user priorities, manages tough assignments, and adapts to personal tastes.
Connected Services and Capabilities
Gemini's support pages list services it links to for fulfilling requests and improving answers. These cover Google Workspace tools such as Gmail, Calendar, Docs, Drive, Keep, and Tasks. Other connections include GitHub, Spotify, YouTube Music, Google Photos, WhatsApp, Google Home, and Android features.
Current documentation outlines actions with varying user effects. These range from pulling data from Workspace apps to setting up calendar events, dispatching messages, launching applications, and operating devices or smart home setups.
Google maintains a Gemini Privacy Hub for context on these links. It covers Google apps and outside services. People can check and erase Gemini Apps Activity, adjust auto-delete options, and decide if data trains Google AI. Users also control app access, data sharing, and details saved through Gemini requests.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Emphasis on Controls and Safety
Google Research outlines standards for AI agents. They require clear human oversight, restricted abilities, visible operations, and planning skills. Google Cloud adds that agent work needs transparency via logs and precise action descriptions. Guidelines stress matching powers to goals and user risk levels, following least-privilege rules.
Remy's ability to learn preferences highlights memory management. The Privacy Hub lets users handle saved information from Gemini interactions and personalization from chat history or Personal Intelligence.
The report lacks specifics on Remy's build, underlying model version, or independence level. Questions remain on whether it operates without user okay and how it manages approvals or action records.
Project Context and Comparisons
The internal paper labels Remy a dog-fooding effort. Tech firms use this term for staff testing before wider rollout. Business Insider draws parallels to OpenClaw, an agent noted earlier this year for independent message responses, user research, and self-directed steps.
OpenAI CEO Sam Altman announced in February plans to hire OpenClaw's maker. Google DeepMind CEO Demis Hassabis has spoken before about creating a digital helper. Google has not verified if Remy will join public Gemini options.
Google, a leader in search and AI since its 1998 founding, acquired DeepMind in 2014 to boost machine learning. Gemini launched in late 2023 as a multimodal model rivaling systems like GPT-4. DeepMind, under Hassabis since 2010, pioneered advances in protein folding and game AI. OpenAI, founded in 2015, competes fiercely, with Altman steering shifts toward agent tech. These tests signal a push toward practical AI assistants amid rising focus on safe, controllable systems.
Related on Neura Market