Meta releases Muse Code, a terminal-based AI agent for large codebases

Meta released Muse Code, a terminal-based AI agent designed to assist developers with complex tasks across large software repositories. The agent is currently available in beta.

Muse Code is powered by Meta’s previously released Muse Spark coding model and can execute complete software engineering tasks, including planning changes, writing code, and validating results.

To handle large projects, Muse Code launches sub-agents that operate in parallel within isolated worktrees, preventing collisions with the user’s working copy. Meta reported internal testing where the agent simultaneously built six features for a game without conflicts.

The agent can be installed with a single command, lowering the barrier to adoption for developers.

Meta’s AI chief, Alexandr Wang, emphasized the cost advantages of Muse Code compared to competitors like OpenAI’s Codex and Anthropic’s Claude Code, positioning it as a more affordable option for certain workflows and use cases.

Meta’s move follows its expansion beyond advertising-focused AI into enterprise markets, including a June announcement of an AI agent for customer service and support.