How to Use Muse Spark 1.2

Muse Spark 1.2 launched on August 5, 2026, through Meta's Muse Code CLI agent. The quickest way to try it is to install Muse Code and run your first command.

This guide walks through installation, browser-based authentication, and the practical differences between using Muse Spark 1.2 in Muse Code and accessing it through the raw API.

Every step is based on Meta's launch materials, and anything not covered in the vendor's documentation is clearly noted.

Access Paths for Muse Spark 1.2

The primary access path is Muse Code, Meta's CLI coding agent that runs Muse Spark 1.2 as its backing model.

Direct API access details beyond Muse Code are not fully documented at launch — verify at https://developer.meta.com/ai/products/muse-code/.

For non-coding workloads, the older Muse Spark 1.1 remains available through the Meta Model API. There is no sibling routing decision inside Meta's own catalog — Meta ships a single Muse Spark line.

• Muse Code CLI: primary access path
• Direct API: not fully documented at launch
• Muse Spark 1.1: still available for non-coding tasks
• Meta developer platform is the source of truth

Install Muse Code

Install Muse Code with a single curl command: curl -fsSL https://dev.meta.ai/install.sh | bash. This is the exact command from Meta's launch email.

The install script drops the muse binary into your PATH. First run triggers browser-based authentication against your Meta developer account.

For CI or headless environments, browser auth will need a workaround — Meta has not documented a headless flow at launch. Verify at https://developer.meta.com/ai/products/muse-code/.

• One-line install: curl -fsSL https://dev.meta.ai/install.sh | bash
• First run: browser auth against Meta developer account
• Headless / CI auth: not documented at launch
• Verify script source before running in production

Your First Muse Code Session

After install and auth, cd into a repo and run muse. The agent reads context from the current directory and spins up worker + reviewer processes in parallel.

First task should be small: ask it to add a missing test for one function. That gives you a full round-trip — plan, edit, run test, event-log entry — inside a few minutes.

The event log is written per session and can be replayed with muse replay. Export the full trace if you need to review or share it with a teammate.

• cd into a repo, run muse
• Start with one small, verifiable task
• muse replay walks the session step by step
• Export the trace for review or handoff

Picking Standard vs Contributor

Pick Standard tier by default. Meta does not use Standard prompts or outputs to improve their products, which is the only defensible posture for client code and PII.

Pick Contributor only for personal projects, OSS contributions, and non-sensitive prototypes. The 12x cost cut is real, but Meta uses those prompts and outputs.

When we work with HOA and condo boards on any AI tool, we keep the tier decision in a written policy — not a per-developer preference. Do the same before onboarding a team.

• Standard by default for anything client-facing
• Contributor for OSS, prototypes, and personal work
• 12x cost cut in exchange for data-use
• Codify the tier decision in a written policy

How to Use Muse Spark 1.2

To use Muse Spark 1.2 effectively, start by choosing an interface that exposes the exact model you need for your workload. Test a new configuration on a single representative task, such as drafting a routine email or parsing a short document, before expanding to full automation. Inspect the saved output directly to confirm formatting, accuracy, and adherence to your prompt instructions.

For coding tasks, Cursor serves as an editor option when the target model appears on its supported list. Check current model availability before setup to confirm the exact model is listed. Once connected, save your prompts and review the generated diffs locally before applying changes to your main codebase.