Inspiration or copying: Meta admits borrowing OpenClo files to run its smart agent Muse

The striking similarity between Meta's smart agent "Muse" and the open-source project "OpenClo" is no longer mere speculation among early users, as the product team at Meta's Superintelligence Labs confirmed that its new agent directly drew its architecture and design from the open project, after developers observed an almost complete match in the internal configuration and runtime files of the two systems.

The issue erupted after a technical dialogue posted by Ansh Nanda, co-founder of an AI application, in which Muse itself responded, describing the overlap between its system files and OpenClo's files as intentional rather than accidental. Researchers and developers soon documented a matching file in Muse that bears the same name used in OpenClo, a markdown-formatted text description known as "sol.md", the core file that configures the agent’s personality, communication tone, values, behavioral limits, and expertise levels, with its contents and environment labels closely aligning between the two models.

The acknowledgment did not come to deny copying, but to justify its adoption as a viable engineering choice.Nat Friedman, Meta’s head of product at Superintelligence Labs, clarified in an official post that Muse was built programmatically from scratch, yet its product was deeply inspired by OpenClo. Friedman, who previously led GitHub before joining Meta, disclosed that he purchased hundreds of Mac Mini machines for his engineering team after testing OpenClo early in the year, considering the project's creator Peter Steinberger, whom OpenAI recruited earlier this year, to have delivered a pioneering and fully sound architecture for framing the work, prompting Meta to emulate it in order to offer a secure, scalable agent for billions.

This move repeats Meta’s historical pattern of absorbing emerging innovations and repackaging them into widely adopted consumer interfaces, the same approach it previously used to clone features from competing apps. The episode gains particular significance as Muse topped the U.S. app store and launched at a faster pace than the early days of ChatGPT, demonstrating that the agents’ battle is decided not only by open architectural innovations but also by the ability of large corporations to secure and mass-distribute the product.

This shift directly impacts development and software engineering teams in the Gulf, Egypt, and the broader Arab region. The development shows technology leaders that the strength of smart agents does not rely solely on massive language-model weights, but fundamentally on the quality of instruction files, behavior specifications, and operational boundaries such as profile files. Rather than awaiting costly, closed-source solutions, technical teams can now examine open frameworks and leverage their structures to build local agents that meet compliance and privacy requirements in Arab enterprises, focusing on turning those structures into secure, scalable, and easy-to-use tools for real-world work environments.