Meta finally shipped the AI agent it has been circling for years, and the internet immediately noticed something familiar about it. Muse looks, walks, and quacks exactly like OpenClaw, the open-source agent that developers have been self-hosting since January. The difference is that Muse has a Meta logo, an App Store page, a couple of hundred thousand downloads a week, and a company that just came out and admitted the whole thing.
Muse launched on September 8 as what Meta called the world's first personal AI agent for everyone. It arrived in the United States as a standalone app and through WhatsApp, ready to handle the boring parts of adult life: booking trips, hunting deals, tracking expenses, and untangling scheduling conflicts.
The numbers arrived fast. According to The Information, Muse pulled in more than 500,000 users in its first week, with more than 250,000 daily active users entering over two million prompts. It hit number one on Apple's App Store with more than 31,000 ratings and topped Google's Play Store. Reports put downloads past 900,000 in six days, beating ChatGPT's early mobile launch numbers.
Déjà Vu in the Workspace
The celebration lasted about as long as it took developers to open Muse's workspace directory. The layout looked eerily familiar to anyone who has run OpenClaw, the viral open-source personal assistant created by Peter Steinberger. Filenames matched. Default content matched. Even the onboarding language appeared to be lifted almost verbatim.
Specifically, the file lineup reads like a shopping list from the OpenClaw repo:
- SOUL.md, the file that defines an agent's personality and rules of behavior
- USER.md, the profile document for the human in the loop
- MEMORY.md, the persistence layer where the agent stores what it knows
- TOOLS.md, the registry of what the agent is allowed to touch
Side-by-side screenshots went viral across X, GitHub, and Hacker News. Search "Muse is literally OpenClaw" and you find AI app co-founder Ansh Nanda's viral post: Muse is LITERALLY OpenClaw for normies. Mocking replies pointed to identical markdown headers and matching memory-file logic. For many observers, it was too close to be coincidence.
Meta held out for about 48 hours of mounting pressure before responding through an engineering lead, and then product chief Nat Friedman took it further on X.
From "Genius" to "Heavily Inspired"
Friedman, the former GitHub CEO who now heads product at Meta Superintelligence Labs, did not do the normal corporate thing. He confirmed that Muse was definitely heavily inspired as a product by OpenClaw, insisted it was built from scratch, and explained that his team fell in love with OpenClaw and set out to build something similar that could scale to billions of people. He also called Steinberger a genius.
When someone asked why the filenames and contents were nearly identical, Friedman's answer was disarmingly simple: we thought that Peter got those things exactly right. It is the rare copycat confession that works better as a compliment than a defense.
Steinberger confirmed that Meta reached out after the findings went public. He stopped short of accusing the company of a license violation, since OpenClaw is permissively licensed, but he expressed frustration that a trillion-dollar company would copy user-facing files without attribution or collaboration. He said he would welcome Meta contributing upstream rather than simply absorbing community work.
TechCrunch reached for the obvious parallel: this is Meta's playbook of copying promising products' best features, same as it did with Snapchat's Stories. Legal experts note Meta is probably safe on copyright grounds, since ideas and functional filenames are hard to protect. Reputationally, the damage is stickier, especially now that the deeper fear has a face: Big Tech treating open-source labs as free research and development.
Security Theater, With a Side of Receipts
Meta's answer to the "you copied the homework" wave is a safety story. The company says Muse runs in a dedicated secure virtual machine, with a separate Sentinel agent that approves outbound actions, and users get an audit trail of what their agent did. At Meta Connect this week, Zuckerberg announced a coming Muse Confidential VM designed so that even Meta cannot see inside.
Trust baggage is real, though. Meta paid $5 billion over Cambridge Analytica. Reports say Amazon restricted Muse's access to its store, citing unauthorized access and security concerns. And the tradeoff cuts against OpenClaw's core appeal: OpenClaw is open and runs on your hardware, while Muse asks you to hand your calendar, wallet, and passwords to a hosted agent wearing a friendly avatar.
So, should you hand Muse your life? The honest answer is a checklist, not a verdict:
- You want zero setup and instant cloud power: Muse is frictionless where OpenClaw is fiddly
- You care about local control and open code: OpenClaw still wins on principle
- You trust Meta with sensitive data: history says ask your lawyer
- You want to watch a trillion-dollar company explain its homework: either app works
OpenAI is reportedly already discussing building its own personal assistant in response to Muse, per The Information. Zuckerberg called Muse the centerpiece of Meta's vision, a future personal super intelligence for billions of people. Meta plans to monetize it by taking a small cut of transactions made through the agent, and it is even shipping a handheld Muse Charm for people who feel weird talking to their phone.
So the real story is not that Meta copied OpenClaw. It is that a solo developer built something so good that the largest social network on Earth decided the only way to beat it was to clone it, rebrand it, and put it at the center of its entire future. Gratitude looks different at scale. Steinberger built a lobster; Meta built the biggest lobster tank on the app store.
Comments