Use the official installer for your platform
The official Code overview provides installers for macOS/Linux and Windows. Read the installer instructions at the source before executing them, then check the installed binary with muse --version. This guide links to the provider rather than mirroring an executable.
Open official Muse Code installation instructions ↗
Start in a small repository with a clean baseline or a recoverable worktree. On first run, inspect the workspace trust prompt and choose your authentication route. A useful first task is a small, reviewable edit with an existing test command, rather than a broad rewrite.
Keep sign-in and billing aligned
The CLI offers browser sign-in or an API key. The subscription documentation explains which onboarding credential receives Code subscription benefits; additional API keys are billed separately. If usage seems to bypass a plan, verify the credential and payment route before assuming a pricing error.
Read the existing Code plan comparison for allowances and plan management. Plan multipliers describe usage relative to another plan; they do not publish a fixed per-token entitlement for your application.
Check the model actually selected
The Code overview’s first-run section names Spark 1.2 as its default, while the broader API docs recommend Spark 1.3 for new examples. These statements refer to different contexts. Inspect the model shown in your installed Code session and current configuration rather than assuming a default from a tutorial.
Regular Wonder and Full are also distinct. A high reasoning setting on regular Wonder is not Full access. Check the Full availability guide before interpreting a model menu or a reported benchmark.
A controlled first repository task
- Record the CLI version, active model, authentication route and baseline commit.
- Ask for a narrow task with explicit acceptance criteria and a known test command.
- Review requested shell actions and permission scope before approval.
- Inspect the diff, run the relevant checks and compare against the original request.
- Record time, retries and billable usage where available; do not turn a single run into a general benchmark.
Code’s official documentation describes approvals and an OS sandbox, with platform differences. Those safeguards are part of the workflow. Installation alone is not evidence that every tool, workflow feature or model is available on every build.
Sources & evidence
Editorial check: 17 October 2026. Product facts come from the following primary sources; planning checklists and examples are our explanations. No paid inference, model benchmark or regional checkout was run for this guide.