Google is racing to build its own answer to OpenClaw — and it goes by the name of Remy.
According to an internal document seen by Business Insider, Google has been quietly developing a new AI “personal agent” codenamed Remy that runs inside a staff-only version of the Gemini app. The agent is described as a “24/7 personal agent for work, school, and daily life, powered by Gemini” — and it’s designed to do far more than answer questions.
“Remy elevates the Gemini app into a true assistant that can take actions on your behalf — not just answer questions or generate content.” — Internal Google document, per Business Insider
What Makes Remy Different
The description paints a picture of an agent that goes well beyond what Google has shipped so far. While Google has been gradually rolling out “Agent Mode” features in Gemini — multi-step task execution with varying levels of autonomy — Remy sounds like a fundamentally different product:
| Feature | Gemini Today | Remy (Internal) |
|---|---|---|
| Interaction | Reactive (user asks, AI responds) | Proactive (monitors, plans, acts) |
| Availability | Per-session | “24/7” persistent |
| Integration | Limited Google services | “Deeply integrated across Google” |
| Learning | Session-based context | “Learns your preferences over time” |
| Autonomy | User-initiated tasks | “Handles complex tasks proactively” |
The most striking phrase in the internal description is that Remy can “monitor for things that matter to you.” This implies a persistent agent that watches email, calendar, documents, and other Google services — and takes action without being explicitly asked.
The Proactive Agent Paradigm
If Gemini today is a search engine that can also write code, Remy is more like a digital executive assistant — one that doesn’t wait for instructions. The document suggests Remy can:
- Monitor your inbox and take action on recurring patterns
- Track project deadlines across Google Workspace and proactively reschedule
- Handle routine administrative tasks (booking travel, filing expenses, managing subscriptions)
- Learn your preferences over time and adapt its behavior
This “proactive agent” paradigm is the direction every major AI lab is racing toward. OpenAI has GPT-5.5 Instant with agentic capabilities, Anthropic’s Claude has computer use, and OpenClaw’s meteoric rise — now at 368,000+ GitHub stars — has proven that there’s massive demand for autonomous digital workers.
Google’s OpenClaw Problem
Google has a unique challenge in the agent space. The company arguably has the best raw ingredients for a world-class personal agent:
- Gemini — competitive with GPT-5 and Claude 4 on reasoning
- Google Workspace — email, calendar, docs, sheets, meet — the full productivity stack
- Android — the world’s most popular mobile OS, with deep system-level access
- Google Cloud — enterprise infrastructure and Vertex AI
- Search, Maps, YouTube, Photos — the richest personal data ecosystem on Earth
Yet Google has no widely available, fully autonomous agent product that competes with OpenClaw, Claude Code, or even the OpenAI Agents SDK. Project Mariner (announced in April 2025) demonstrated browser-based agent capabilities but never reached public release. The “Agent Mode” features in Gemini are useful but limited.
Remy appears to be Google’s attempt to solve this by going bigger than a browser extension — building an infrastructure-level agent that lives inside Google’s ecosystem and acts as a persistent digital proxy for the user. For more on Google’s agent strategy, see our complete guide to AI agents and the coverage of DeepMind’s AlphaEvolve.
The Ratatouille Connection
The name “Remy” is almost certainly a reference to the Pixar film Ratatouille — about an unlikely chef who turns out to be a rat. As one HN commenter noted, it’s a curious choice for a company that has historically avoided naming AI products after things that evoke "hidden" or "unseen" actors. For more on the personal AI agent landscape, see our state of AI agents May 2026.
The Dogfooding Phase
Remy is currently being tested by Google employees in a “dogfooding” project — standard practice at tech companies where employees use pre-release products to surface bugs and gather feedback.
The timing is notable: Google’s I/O developer conference is later this month, where the company is expected to showcase its next wave of AI products. Agents will almost certainly be a major focus. Google DeepMind CEO Demis Hassabis has long talked about his vision to build a truly capable digital assistant, and Remy may be the first concrete expression of that vision.
Competitive Landscape
Remy enters a market that’s rapidly heating up:
- OpenClaw — the open-source phenomenon now at 368,000+ stars, with autonomous capabilities across messaging, voice, file management, and recently Google Meet integration
- Anthropic’s Claude — expanding into enterprise financial services agents and computer use capabilities
- OpenAI — with GPT-5.5 Instant and CopilotKit raising $27M for app-native agent deployment
- Meta — launching Muse Spark from its new Superintelligence Labs
- AWS — enabling agents to drive Workspaces virtual desktops
What differentiates Remy is depth of integration. OpenClaw connects to services through APIs and browser automation. Remy would theoretically have native, privileged access to the entire Google ecosystem — email parsing, calendar introspection, document editing, search history — the kinds of deeply personal data that enable an agent to act with genuine contextual understanding.
Privacy and Trust Questions
A 24/7 proactive agent that monitors your digital life raises obvious privacy considerations. Google has faced intense scrutiny over data collection practices, and a persistent agent with read-write access to email, calendar, and documents would represent a significant escalation.
The internal document doesn’t address privacy guardrails, but any public launch would require:
- Granular permission controls (which services the agent can access)
- Human-in-the-loop oversight for consequential actions
- Transparent audit trails of agent activity
- Easy revocation — “fire the agent” should be a single button press
- On-device processing where possible to minimize data exposure
What’s Next
Google remains tight-lipped about Remy’s public launch timeline. The internal document describes it as a “dogfooding project,” which suggests months of internal testing before any public release. Given the I/O conference later this month, a developer preview or technology demonstration seems plausible.
One thing is clear: the consumer agent race is no longer a sideshow. With OpenClaw demonstrating viral demand, Cloudflare building the infrastructure for agent-provisioned services, and now Google investing in a first-party personal agent, the question is no longer if autonomous AI agents will become part of daily life — but which company will be the one we trust to run them.
Google has the data, the distribution, and now the product vision. The question is whether they can execute before the open-source wave makes proprietary agents irrelevant.