Meta has entered the AI coding agent market with Muse Code, a Terminal-based tool for macOS that targets complex software engineering tasks across large repositories. No app interface, no GUI. You install it with a single command and work through the Terminal.
What It Does
According to Meta’s announcement, Muse Code handles planning changes, writing code, validating results, and coordinating multiple persistent subagents simultaneously. The pitch is transparent operation across large codebases, which puts it in direct competition with Claude Code and ChatGPT Codex.
The absence of a graphical interface is worth flagging. For developers comfortable in the Terminal that’s a non-issue. For anyone who relies on a visual workflow, it may be a barrier compared to alternatives that ship with an app.
Install Command
curl -fsSL https://dev.meta.ai/install.sh | bashThat’s the full install. Paste it into Terminal, run it, and you’re on the contributor tier.
Pricing
Muse Code uses pay-as-you-go pricing tied to token consumption:
- Input tokens: $1.25 per million
- Output tokens: $4.25 per million
No flat subscription tier was mentioned at launch. Costs scale directly with usage.
The Model Behind It
The Muse Code launch was paired with the release of Muse Spark 1.2, a model Meta says was trained specifically for coding assistance. Training included rejection-sampled harness trajectories and recipe optimizations for goals, compaction, and subagents.
One Caveat Worth Noting
The source article flags Meta’s data history as a real consideration before installing software that runs with access to your codebase and local environment. Meta has faced significant regulatory action over data practices in the past. Whether its AI tooling operates under different safeguards is not addressed in the announcement, and that’s a gap worth evaluating before you pipe a script from Meta’s servers directly into bash.
