Meta admits Muse was built under OpenClaw’s influence

What Meta borrowed

The comparison began with a viral post by Ansh Nanda, co-founder of an AI app developer. Nanda called Muse “literally OpenClaw for normal users” and attached a conversation with the agent.

Muse reportedly said the similarities in its system files and OpenClaw’s were not accidental. Users then began comparing the two systems themselves. One found that Muse included SOUL.md, a Markdown configuration file that defines:

Another user found that the overlap extended beyond file names: the contents of SOUL.md in the two assistants were almost identical.

Friedman did not dispute the comparison when asked why Muse used the same workspace file names and nearly identical content. Instead, he said Meta’s team considered OpenClaw creator Peter Steinberger’s decisions carefully judged.

That response is more significant than a denial would have been. Meta is not presenting Muse as an unrelated invention that happened to converge on the same structure. It is acknowledging that OpenClaw supplied much of the product’s conceptual starting point.

The product Meta wants to scale

Friedman said he encountered OpenClaw in January and bought hundreds of Mac mini computers for the Meta Superintelligence Labs team. Many employees, he added, became fans of OpenClaw and other personal AI agents.

Friedman called Steinberger a genius and said the agent wrapper he created was innovative from the beginning and had inspired many developers. Meta’s objective with Muse was to build something similar, but safe, secure, convenient and usable by billions of people.

That ambition fits Meta’s established product pattern: take a promising product and reproduce its most effective features. The clearest earlier example is the Stories format, which first appeared on Snapchat.

OpenClaw’s open-source distribution also makes the borrowing easier to understand. Its rapid rise in popularity drew attention well beyond its original users, and OpenAI acquired its creator earlier this year.

I think the interesting distinction here is not between “original” and “copied.” It is between an agent designed around a developer community and one redesigned for scale. Meta is betting that the structure people value in OpenClaw can survive a much larger, more controlled product.

Muse has momentum, but the disclosure is thin

Muse recently reached number one in the US App Store. Data also shows that, when comparing the platforms and their market availability directly, Muse is currently ahead of ChatGPT in rollout speed.

Those results give Meta a reason to emphasize adoption rather than provenance. But the announcement leaves the most consequential part unclear: how much of OpenClaw’s underlying behavior, not just its file structure, made it into Muse?

What I’d want to know is where Meta drew the line between inspiration and reproduction. The company has explained why the resemblance exists, but not how extensively the two agents share design decisions beyond the files users can inspect.

A Meta representative told TechCrunch that the company had nothing to add beyond Friedman’s statement. Muse may therefore be succeeding precisely because it packages a recognizable agent model for a broader audience—but its strongest evidence of product differentiation is still missing from the announcement.