Muse Code (Beta) and Muse Spark 1.2

A terminal coding agent for long-horizon software engineering, plus the Muse Spark 1.2 model co-trained with it, for developers on macOS and Linux.

Muse Code plans changes, writes code, and validates results across large multi-file repositories from the terminal
A simple primary agent loop is paired with persistent async background sub-agents that stay alive for the whole session, so they avoid re-gathering context and need less steering on hard tasks
Every model call, tool run, approval, and edit is appended to a local append-only event log, which Meta describes as replay-exact and restart-safe for crash recovery and exact session replay
Ships with three skills: /plan for approval-gated planning, /grill for stress-testing, and /goal for objective completion
Muse Spark 1.2 was co-trained with the agent using a self-improvement loop in which Muse Spark 1.1 generated training data, with scaled-up coding-task compute and environment diversity
Installs on macOS and Linux via curl -fsSL https://dev.meta.ai/install.sh | bash; the model is also on the Meta Model API at $1.25 per million input tokens and $4.25 per million output