Muse code

Muse code is Meta's beta terminal coding agent that runs on macOS and Linux. It helps developers automate complex coding tasks by planning changes, writing code, validating results, and coordinating multiple background agents. Unlike a simple code generator, Muse code manages state across sessions and can resume work after interruptions.

How it works: Muse code uses the Muse Spark 1.2 model for inference but separates the agent runtime from the model. It maintains a local event log that records every action — model calls, tool runs, approvals, and edits. This log makes sessions auditable and restart-safe. A standout feature is its fan-out mode, where write-capable child agents work in isolated git worktrees, allowing parallel work on different parts of a codebase. Meta claims persistent background agents stay active during a session instead of being recreated for each subtask.

Who benefits: Software developers and engineering teams working on multi-file changes, long-running refactoring tasks, or complex debugging sessions. Use cases include automating a series of code changes across a repository, fixing failing tests, and managing large-scale codebase modifications with crash recovery. Since it's in early beta, developers should test in non-critical repositories first.