Meta pushed its Muse Code coding agent out of beta on August 31, adding a TypeScript SDK, a multi-agent Workflow orchestrator, and three subscription tiers priced from $5 to $50 a month. Less than a month after the command-line agent first shipped, the update turns Muse Code from a terminal tool into a programmable platform that builders can extend and embed.
What You Can Do With It Now
The headline addition is the Muse Code SDK, in developer preview. It lets you build your own agents on top of Muse Code, drop them into an app, wire in custom tools, stream progress, and resume sessions later, all from TypeScript. Sessions run on the Muse Session Protocol, an open standard-I/O protocol with no server and no network, so the whole loop executes on your machine.
The new Workflow orchestrator spins up persistent sub-agents that run in parallel on isolated copies of your project, so you can fan a large refactor out across several agents at once and watch them from one control room. Inter-session messaging lets those agents pass warnings and answers to each other, and a double-Esc Rewind rolls a session back to a safe point when an agent goes off the rails. Start from the Muse Code docs and cookbooks.
Why It Matters for Creators
Muse Code is Meta's answer to Claude Code, Codex, and Cursor, and the SDK is what makes it different: instead of using someone else's agent, you can script your own build pipeline and run it locally. That matters right now, because model access keeps getting pulled out from under builders, as when OpenAI wound down GPT access inside Cursor. A local-first, self-hosted agent that you control end to end is harder to de-integrate. The agent first shipped in beta on August 5 running on Meta's Muse Spark model.
Key Details
Pricing. Alongside the flat $5 to $50 monthly plans, usage-based rates run $1.25 per million input tokens and $4.25 per million output, with cached input at $0.15. As The New Stack reported, a contributor tier drops that to $0.10 input and $0.20 output if you let Meta train future models on your prompts and completions.
SDK. TypeScript, developer preview, built on the Muse Session Protocol.
New features. Workflow multi-agent orchestration, inter-session messaging over Unix sockets, and Rewind rollback.
What to Do Next
If you already live in a terminal agent, try Muse Code on a throwaway branch and set up one Workflow that splits a refactor across two sub-agents to see how the parallel model fits your work. If you build tools, the SDK is the piece to test first: prototype a small custom agent, decide whether the contributor tier's data-sharing tradeoff is acceptable for your codebase, then price it against your current Claude Code or Codex spend before switching anything in production.
