Meta’s Muse Code arrives on macOS
A one-line command in your macOS terminal can get you access to Meta’s Muse Code, which aims to be a transparent AI coding tool that can work across large repositories.
There are already coding agents like Claude Code and ChatGPT Codex available for macOS, but Meta wants to enter the ring with its latest AI toolset. Unlike the others, it doesn’t have an app interface and runs through the Terminal.
According to an announcement post from Meta, Muse Code can take on complex software engineering tasks across large repositories. It is capable of planning changes, writing code, and validating the results while coordinating multiple persistent subagents.
As always with these AI announcements, remember that the language being used attempts to anthropomorphize the tools to be more human-like. Nothing about an AI behaves like a human — they don’t think, reason, organize, sympathize, or understand.
AI is a tool, and if Muse Code sounds like it could help you develop software, it might be a particularly useful tool. Developers have shared just how useful having a coding agent available can be, especially for troubleshooting bugs.
To install Muse Code, you’ll just need to paste in a single command into Terminal:
curl -fsSL https://dev.meta.ai/install.sh | bash
Of course, this isn’t free. Those that install Muse Code will start on Meta’s contributor tier. Developers join with a pay-as-you-go option which charges $1.25 per million tokens in input and $4.25 per million tokens of output.
The Muse Code announcement was paired with the release of Muse Spark 1.2, which was trained with coding assistance in mind. Meta says the training included rejection-sampled harness trajectories and recipe optimizations for goals, compaction, and subagents.
Choose your AI partner
Meta’s inevitable introduction of a coding agent was a long time coming. The company has been poaching Apple employees in its expensive chase for so-called superintelligence.
Muse Code is also another revenue stream to cover the incredible costs of AI and research around it. Meta previously tied some of its premium subscriptions to AI tool use as well.
There are several AI coding tools to choose from, so it will be interesting to observe if Meta can make a dent in the market. Not having a direct UI and interfacing via Terminal could be a problem for some.
Then there’s the privacy aspect. Meta is known for its horrific history of scraping every piece of data it can from users, and while it might have safeguards in place for its AI tools, it seems unwise to trust Meta software installed with access to so much information.