AI Pulse · Frontier Labs Watch

Meta has moved Muse Code out of beta and added two things that matter for companies building internal AI tools: a TypeScript SDK in developer preview, and monthly subscription tiers alongside pay-as-you-go pricing.

The SDK is designed for teams that want to build their own agents on the Muse Session Protocol. The notable part is that it runs fully locally, with no server required. In practical terms, that could suit businesses that want more control over where an agent runs, how it connects to internal systems, and what data leaves an employee’s machine. It does not mean every deployment will be simple. A local setup still needs clear access controls, device management, and a plan for maintaining the tools people depend on.

The new pricing options also signal that Meta is trying to support more than experimental use. Pay-as-you-go works when usage is uncertain. Monthly plans can make budgeting easier once a team has regular users or repeatable workflows. There is also an existing contributor option tied to opting training data in, so owners should read those terms carefully before choosing a lower-cost route.

For operators, the question is not whether to adopt another coding tool. It is whether local AI agents could handle specific internal work without creating another unmanaged software layer. This is the kind of thing we build into an AI command centre, with visibility over what agents are doing, which systems they can access, and where the business gets value.