Introducing Muse Code, a terminal coding agent powered by Muse Spark 1.2, with persistent background agents, repository-scale execution, and built-in verification.
Loading comments…
Maker
kettle_dev
Visit Website
research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2
Project Info
Product Keywords
Alternatives
Muse Code is a terminal coding agent powered by the Muse Spark 1.2 model, designed to handle complex software engineering tasks across large repositories. It plans changes, writes code, and validates results—all from the command line on macOS or Linux. The tool pairs a straightforward agent loop with persistent background agents, making it capable of tackling multi-step problems with less manual steering and greater accuracy.
Muse Code keeps specialized background agents active throughout an entire session, rather than spawning them per task. These agents carry out next steps independently and decide when to report back, which cuts down on redundant information gathering and reduces latency on difficult, multi-step work.
Every model call, tool run, approval, and edit is appended to a local event log. This single source of truth makes the runtime replay-exact and restart-safe, so after a crash the agent resumes precisely where it stopped—ideal for long-running tasks that can't afford to start over.
Muse Code ships with three default skills: /plan converts a task into an approval-gated plan, /grill stress-tests that plan until it holds up, and /goal drives the agent toward successful completion of the specified objective. These built-in commands give you structured control without extra setup.
The underlying model was co-trained with Muse Code itself, using rejection-sampled harness trajectories and recipe optimizations for goals, compaction, and subagents. This tight coupling ensures the model performs at its best specifically when driving the Muse Code toolset.
Muse Code doesn't just write code—it persists through failures, coordinates subagents, and validates its own work across entire repositories.
Other tools you might consider
The combination of persistent background agents and a crash-safe event log is rare in terminal coding tools. Most agents lose context or restart from scratch on failure, but Muse Code's replay-exact runtime means long-horizon tasks—like the kernel optimization case study with 1,000+ tool calls over 24 hours—stay on track. The bundled /grill skill also adds a stress-testing layer that pushes plans until they're solid, which goes beyond typical code generation.
You're a developer who lives in the terminal and regularly faces large, messy repositories where a single mistake can cascade. If you've been frustrated by agents that lose context, need constant hand-holding, or can't survive a crash mid-task, Muse Code's persistent subagents and restart-safe design directly address those pain points. It's also a strong fit if you're working on performance-critical code—like GPU kernels—where iterative compile-profile-improve loops demand an agent that can sustain focus over hours without derailing.
