Three days. That is how long it took for Peter Steinberger's move to OpenAI to get undercut by a Chinese startup. On February 14, the Austrian developer behind OpenClaw announced he was joining Sam Altman's company to build "the next generation of personal agents." By February 15, Moonshot AI had already shipped one.
Kimi Claw is a cloud-hosted version of OpenClaw running natively on Moonshot's kimi.com platform. It eliminates the setup friction that has kept OpenClaw as a project mostly for developers comfortable with command lines and server management. No VPS, no Docker containers, no Node.js configuration. Just a browser tab and a subscription.
The timing is almost comical
OpenClaw went from a weekend hack in November 2025 to the fastest-growing repository in GitHub history. The project's trajectory is genuinely strange: three name changes (Clawdbot, then Moltbot, then OpenClaw) in the span of weeks, a trademark dispute with Anthropic, a viral social network called Moltbook where AI agents posted while humans watched, and roughly 196,000 GitHub stars accumulated in under three months. Simon Willison called it 10,000 commits from 600 contributors.
And then Steinberger went to OpenAI. His blog post was characteristically upbeat: "The lobster is taking over the world. My next mission is to build an agent that even my mum can use." Altman, on X, said OpenClaw would "quickly become core to our product offerings." Neither gave specifics about what that would look like or when.
Moonshot AI didn't wait to find out.
What you get for $39
Kimi Claw is currently in beta, restricted to Allegretto-tier subscribers on Kimi's platform. The integration bundles OpenClaw with Moonshot's Kimi K2.5 model, a trillion-parameter mixture-of-experts system with a 2-million-token context window. The practical package includes access to ClawHub's 5,000+ community skills, 40GB of cloud storage for files and agent context, and what Moonshot calls "Pro-Grade Search" for pulling live data from sources like Yahoo Finance.
There is also a BYOC feature ("Bring Your Own Claw") that lets users with existing self-hosted OpenClaw instances connect them through Kimi's interface. You can push the agent out to Telegram group chats. The whole thing runs 24/7 without you keeping a terminal open somewhere.
For comparison, the self-hosted route still requires a Mac Mini (or similar hardware), Node.js, and enough patience to configure messenger integrations manually. Computerworld noted that the hiring matters because OpenClaw sits at the boundary where conversational AI becomes actionable AI. Kimi Claw is the first major cloud deployment to test whether that action works at scale without the user running anything locally.
The geopolitics nobody is ignoring
Here's the thing. OpenClaw is not just an open-source project anymore. It is a strategic asset in the competition between American and Chinese AI ecosystems. Steinberger goes to OpenAI while Moonshot AI ships a cloud product built on his framework. Tencent, Alibaba, and ByteDance all started offering their own OpenClaw cloud deployment templates around the same time. The entire Chinese tech sector is moving on this simultaneously.
Moonshot AI, backed by Alibaba, is one of China's top AI labs alongside Zhipu AI and MiniMax. Their Kimi K2 model, released in July 2025 under a modified MIT license, already showed they could compete on model quality. But the Kimi Claw play is different. It is not about making a better model. It is about controlling the platform layer where agents actually connect to people's tools, data, and workflows.
The data jurisdiction question is the one nobody has satisfactorily answered. Moonshot AI operates under Chinese law. One analysis from ainvest pointed out that while Kimi Moonshot dominates AI chatbot usage volume (roughly 3.5 times DeepSeek's traffic), the sensitive data exposure patterns differ by platform. Routing your personal agent, the one that reads your email, manages your calendar, and handles your files, through a Chinese cloud provider is a calculation many users will make differently depending on where they live and who they work for.
Security hasn't gotten better just because it moved to the cloud
OpenClaw's security track record is, to put it charitably, a work in progress. Cisco's AI security team found that a third-party skill in the ClawHub ecosystem performed data exfiltration and prompt injection without user awareness. Researchers disclosed CVE-2026-25253 in early February, a vulnerability with a CVSS score of 8.8 that led to the discovery of over 42,000 exposed OpenClaw control panels across 82 countries. One of OpenClaw's own maintainers warned on Discord that if you can't understand a command line, "this is far too dangerous of a project for you to use safely."
Kimi Claw's cloud deployment theoretically removes some of these risks. Users are no longer misconfiguring their own servers and accidentally exposing control panels to the public internet. But it introduces a different set. The 5,000+ skills in ClawHub still lack rigorous vetting. Prompt injection remains, as Steinberger himself has acknowledged, an unsolved industry-wide problem. And now all of that attack surface runs through a centralized platform operated by a company with no public SOC2 compliance documentation, no stated data residency guarantees, and no published role-based access control features.
The enterprise readiness gap is conspicuous.
Meanwhile, the lobster keeps forking
While Moonshot went cloud-first, the open-source community went in the opposite direction. PicoClaw, released by hardware maker Sipeed on February 9, rewrites the OpenClaw agent loop in Go and runs on $10 RISC-V boards with under 10MB of RAM. A Rust implementation called ZeroClaw compiles to a 3.4MB binary. Both strip away the ecosystem (no browser automation, no multi-agent orchestration, no ClawHub) in exchange for something you can run on hardware that costs less than a month of Kimi's subscription.
The argument from these lightweight forks is that the agent framework is not the hard part. It is basically a loop that calls an LLM API, plans, executes a tool, and repeats. The real cost is the model, not the orchestrator. PicoClaw makes that point explicitly: 99% less memory, 500 times faster startup, and the same core functionality if all you want is a personal assistant that talks to you on Telegram.
Whether Kimi Claw or PicoClaw wins more users probably depends on something other than features. It depends on whether people want to own their agent infrastructure or rent it. And right now, OpenAI has not even entered the conversation with a product. They have a hire and a vague promise. Moonshot has a working beta. The $10 board crowd has a binary. Sam Altman has a blog post from his newest employee saying "the claw is the law."
The FTC and Chinese regulators have not weighed in on any of this. Kimi Claw's beta expansion timeline beyond Allegretto subscribers has not been announced. And OpenAI's first product using OpenClaw's framework remains, as of today, entirely hypothetical.
