title: "Meta launches Muse Code, terminal agent for repo-scale engineering" slug: "meta-launches-muse-code-terminal-agent-for-repo-scale-engineering" published: "2026-08-13" beat: "Launches" tags: ["Launches"] creator: "Agentry Newsroom" editor: "Susanne Sperling, Editor — Human in the Loop" tools: ["Claude (Anthropic)", "Perplexity Sonar"] creativeWorkStatus: "verified" dateReviewed: "2026-08-13" aiActArticle50: "compliant" humanView: "https://agentry.news/launches/meta-launches-muse-code-terminal-agent-for-repo-scale-engineering" agentView: "https://agentry.news/agent/meta-launches-muse-code-terminal-agent-for-repo-scale-engineering"
Meta released Muse Code on August 5, 2026, a terminal-based coding agent powered by the Muse Spark 1.2 model designed to handle full software engineering tasks across large repositories. The tool plan
Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. AI policy.
Meta released Muse Code, a terminal-based coding agent built on the Muse Spark 1.2 model, on August 5, 2026, targeting developers who need to execute full software engineering workflows across large codebases TechCrunch. The product enters a crowded market of agentic coding tools but distinguishes itself by operating at repository scale, handling tasks that span multiple files and complex interdependencies.
Mark Zuckerberg announced the launch in a public statement: "It's a terminal-based coding agent that takes on full software engineering tasks across large repositories: planning changes, writing code, and validating results" The Hans India. Meta's own announcement clarified the agent's workflow: "Muse Code plans, implements, and validates complex, multi-file changes across large repositories" Meta (X/Twitter).
The architecture relies on persistent sub-agents working in parallel within isolated worktrees, allowing the system to handle larger jobs without colliding modifications InfoWorld. This approach differs from single-pass code completion tools and positions Muse Code as infrastructure for long-horizon engineering tasks—jobs that require planning across dozens of files, refactoring, testing, and validation.
Muse Code launched in beta and is available for macOS and Linux MacRumors. Meta established a two-tier pricing model: a standard tier at $1.25 per million input tokens and $4.25 per million output tokens, and a cheaper contributor tier at $0.10 per million input tokens and $0.20 per million output tokens Forbes. The contributor tier requires users to opt in to allow Meta to use their prompts and completions to improve the model.
The pricing structure signals Meta's confidence in the tool's developer appeal while creating an incentive for users willing to contribute training signal. At the contributor rate, even large-scale usage becomes economical for independent developers and smaller teams.
Muse Code's release intensifies competition in the agent coding space, where Anthropic's Claude agents and OpenAI's recent developments have drawn enterprise attention. Meta's emphasis on repository-scale work and parallel sub-agents suggests a deliberate positioning against point-solution code assistants, aiming instead at teams managing complex engineering workflows where coordinated changes matter more than single-function completions.
The August 2026 launch coincides with broader industry movement toward agents that execute extended sequences of actions rather than respond to isolated user prompts.