How to Use Meta Muse Code: Setup, Workflow, and First Project Guide
Official install command and cautious first-run workflow for Meta Muse Code, with notes on its published skills, background agents, event log, and product limits.
Published: Last checked:

Official install command and cautious first-run workflow for Meta Muse Code, with notes on its published skills, background agents, event log, and product limits.
Written and checked by the RankLLMs editorial team. See theeditorial policy.
Meta released Muse Code beta for macOS and Linux as a terminal coding agent powered by Muse Spark 1.2. Meta’s announcement describes planning changes, writing code, validating results, and coordinating persistent background agents. It also documents a local event log and bundled skills such as /plan, /grill, and /goal.
This guide sticks to the setup and workflow details Meta has published. The release page is the source of truth for current installation instructions and product availability.
Install Muse Code
Meta’s official announcement gives this installer for macOS and Linux:
curl -fsSL https://dev.meta.ai/install.sh | bashAs with any shell installer, review the script and your organization’s software policy before running it. Follow Meta’s page for any prerequisites or changes. The announcement does not document a Windows installer; if you use Windows, check Meta’s current documentation for supported options rather than assuming a WSL workflow is supported.
Start with a small repository task
Launch Muse Code from a repository you are allowed to modify. Begin with a bounded, read-only request, such as asking it to map the project structure and identify relevant files. Confirm that it did not change files before moving to an edit request.
For implementation, specify the intended behavior, affected areas, constraints, and a verification command that matches the repository’s own package manager and scripts. Ask the agent to summarize changes and show the diff. Inspect that diff yourself and run the checks locally; a model saying a check passed is not a substitute for your own result.
Use the built-in skills and background agents
Meta says Muse Code includes /plan, /grill, and /goal skills. It also describes persistent asynchronous background agents that can carry out work during a session. Exact prompts, permissions, and behavior may vary by release. Use the current product help to confirm how to invoke skills and how approvals work.
Meta says the runtime appends model calls, tool runs, approvals, and edits to a local event log. The announcement describes this log as replay-exact and restart-safe. Treat that as a product claim about recorded runtime state; it does not mean that rerunning a task will produce identical model outputs or that the log cannot contain sensitive material. Check current data handling guidance before using it with private code.
Check before trusting an agent run
- Inspect proposed commands and the files the agent can access.
- Avoid exposing API keys, customer data, or secrets in prompts and logs.
- Review the patch, then run the repository’s tests and checks yourself.
- Keep a clean checkpoint or branch before a broad change.
- Verify that an agent’s tools, approvals, and network access match your team’s policy.
Meta’s release describes a model-and-harness kernel optimization demonstration involving over 1,000 tool calls over up to 24 hours. This was a specific vendor case study using Muse Code and Muse Spark 1.2, not a promise that ordinary tasks run unattended or achieve a particular result.
Muse Spark 1.2 was the model paired with Muse Code at launch. Meta later announced Muse Spark 1.3 with improvements for longer-horizon tasks; check the latest Meta model updates for newer model context. See the original Muse Code release for installation and product details.
Frequently Asked Questions
How do I install Meta Muse Code?
On macOS and Linux, install Muse Code via the official terminal script: 'curl -fsSL https://dev.meta.ai/install.sh | bash'. Restart your shell after installation.
Where do I find authentication instructions for Muse Code?
Follow Meta's current Muse Code onboarding and developer documentation. The launch announcement does not document credential-generation steps or a credential storage path.
Can I run Meta Muse Code on Windows?
Meta's launch announcement documents installation on macOS and Linux. It does not document a Windows installer or guarantee WSL support; check current Meta documentation.
How does Muse Code handle multi-file coding tasks?
Meta says Muse Code can coordinate persistent asynchronous background agents. The release does not promise that these agents use isolated Git worktrees or avoid file conflicts; review concurrent edits and the final diff.
Source links for data and claims discussed in this article are listed below. Refer to each source for its current figures and methodology.
Recommended Reading
Related by topic and recency.
Muse Code vs Claude Code: Pricing & Pilot Plan
Compare Muse Code and Claude Code on documented features and billing, then evaluate both coding agents on your repositories with a matched pilot plan.
Muse Spark 1.2: What Changed in Meta's Coding Model
Meta describes Muse Spark 1.2 as a coding-focused update co-trained with Muse Code, with long-horizon task training, context compaction, and a published kernel optimization case study.
Claude Code vs Gemini CLI vs GitHub Copilot CLI (2026 Tested)
Terminal AI coding agents compared: Claude Code CLI, Google Gemini CLI, and GitHub Copilot CLI tested on multi-file edits, context limits, and pricing.
