Muse Code Explained

Muse Code is Meta's new CLI coding agent, released 2026-08-05 and backed by the Muse Spark 1.2 model. It is a purpose-built coding agent optimized for long-horizon, multi-agentic workflows and transparent auditability.

This guide covers the four feature buckets Meta announced, how to install it, and where it fits alongside Claude Code, Cursor, Codex, and other coding agents.

Everything below is quoted from Meta's launch materials. Anything not published (rate limits, memory backend, full tool list) is called out so you can verify at the vendor page.

What Is Muse Code?

Muse Code is Meta's CLI coding agent that runs Muse Spark 1.2 as its backing model. Meta positions it as "a purpose-built coding agent optimized for long-horizon, multi-agentic workflows and transparent auditability."

It installs with a single curl command and authenticates through the browser against your Meta developer account.

It joins the CLI coding-agent category alongside Claude Code, Codex, Gemini CLI, and Aider — but it is the first to ship multi-agent-by-default as a first-class feature.

• Category: CLI coding agent
• Backing model: Muse Spark 1.2
• Vendor: Meta
• Install: curl -fsSL https://dev.meta.ai/install.sh | bash

The Four Announced Feature Buckets

Meta grouped Muse Code's launch story into four buckets. Every one is quoted directly from the announcement.

Multi-agent by default: "Multiple agents coordinate on every task — workers in parallel and reviewers in the background, so you ship faster without sacrificing quality."

Auditable by construction: "Full control and visibility over everything Muse Code does. Every file edit, tool call and decision is recorded in an event log, so you can replay any session step by step or export the full trace."

• Multi-agent by default (workers + background reviewer)
• Auditable by construction (full event log, replay + export)
• Pick up where you left off (persistent memory + session continuity)
• Competitive pricing (Standard vs Contributor two-tier)

Persistent Memory and Two-Tier Pricing

Pick up where you left off: "Persistent memory and session continuity mean you can close your laptop and come back the next day with no lost work." The memory backend (local vs cloud) is not documented at launch.

Muse Code inherits Muse Spark 1.2 token pricing. Standard is $1.25 in / $0.15 cached / $4.25 out per M tokens. Contributor is $0.10 in / $0.002 cached / $0.20 out per M tokens.

Contributor is roughly 12x cheaper in exchange for Meta using prompts and outputs to improve their products. Standard is the default privacy-preserving tier.

• Persistent memory across sessions
• Standard: $1.25 / $0.15 / $4.25 per M tokens
• Contributor: $0.10 / $0.002 / $0.20 per M tokens
• Contributor uses your data; Standard does not

Install and First Run

Install with one line: curl -fsSL https://dev.meta.ai/install.sh | bash. First run triggers browser authentication against your Meta developer account.

cd into a repo and run muse. The agent reads context, spins up worker + reviewer processes, and starts the task loop.

When we work with client teams on any new tool we run one small verifiable task first (like adding a missing test) before committing to a workflow change. Do the same here.

• One-line install (curl)
• Browser auth against Meta developer account
• cd + run muse to start
• Start with one small verifiable task

Where Muse Code Fits in the Category

Muse Code slots in alongside Claude Code, Codex, Gemini CLI, and Aider as a CLI-native coding agent. It is not an IDE plugin.

Its differentiators are multi-agent-by-default (not opt-in), the full event log (not just chat history), and the Contributor pricing tier (best-in-class for non-sensitive work).

For IDE-embedded workflows (Cursor, VS Code Copilot), Muse Code is complementary, not a replacement. Route long-horizon repo-wide tasks to Muse Code; keep quick single-file edits in the IDE.

• Direct rivals: Claude Code, Codex, Gemini CLI, Aider
• Differentiators: multi-agent-default, event log, Contributor tier
• IDE assistants: complementary, not replaced
• Route by task horizon

How to Use Muse Code

To use Muse Code 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.