Meta AI Releases Muse Code (Beta): A Terminal Coding Agent Powered by the New Muse Spark 1.2 Model
Meta's new terminal coding agent plans, writes, and validates across entire codebases — with persistent async workers that survive crashes.
Why it matters
Muse Code represents a shift toward persistent, stateful agents designed for multi-step repository-scale work rather than one-off task completion. The architecture (async background agents, replay-exact event logs, co-trained harness) is a genuine engineering advance in agent reliability and continuity — directly actionable for teams considering autonomous coding in production.
The key facts
6 to knowMeta Superintelligence Labs released Muse Code as terminal coding agent in beta
Powered by new Muse Spark 1.2 model
Agent plans changes, writes code, validates results across large repositories
Async background agents remain active for entire session (not spawned per-task)
Local append-only event log enables replay-exact runtime and crash-safe restart
Muse Spark 1.2 co-trained with harness for long-horizon, repository-scale work
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Meta Superintelligence Labs has released Muse Code, a terminal coding agent in beta, powered by the new Muse Spark 1.2 model. Muse Code plans changes, writes code, and validates results across large repositories. Async background agents stay active for the whole session instead of spawning per…