Meta Platforms released Muse Code in beta on Wednesday, positioning the tool as a terminal-based coding agent capable of carrying out comprehensive software engineering work across large repositories.
Muse Code is engineered to plan changes, write the necessary code and validate the results produced. The agent is powered by Muse Spark 1.2, an updated model Meta describes as focused on coding tasks.
Unlike ephemeral systems that restart for each new request, Muse Code employs specialized background agents that remain active for the duration of a user’s session. These background agents accumulate context over time, rather than rebuilding context each time a fresh task is started. When a task grows beyond a certain scope, Meta’s system launches separate sub-agents that run in parallel. Those sub-agents operate within isolated worktrees so the main working copy remains untouched while concurrent work proceeds.
In internal testing, Meta used the system to develop six features for a game simultaneously and reported that the features were built without conflicts. The company also emphasized ease of adoption: users can install Muse Code with a single line of code and begin using it on a contributor tier. Meta described initial pricing for that contributor tier as low-cost.
The beta release highlights three engineering design choices: a coding-optimized model (Muse Spark 1.2), persistent background processes that build session continuity, and a workflow that isolates parallel sub-agents to avoid interfering with the primary working copy. Meta’s testing example—building six game features concurrently without conflicts—illustrates how the isolation approach is intended to reduce merge friction during parallel development.
For teams evaluating Muse Code, the one-line install and contributor tier pricing are concrete operational and economic details provided by Meta. Beyond those specifics, the company’s statements focus on the system architecture and its behavior under concurrent workloads rather than on broader deployment metrics or long-term pricing expectations.