How Mux Built Durable Video Workflows with the @mux/ai SDK

AI workflows often fail mid-execution due to network timeouts or API limits. In such cases, restarting from scratch causes duplicate costs, or requires developers to write complex code themselves to manage state.

To solve this, Mux developed the @mux/ai SDK and adopted Vercel's Workflow DevKit. The core idea is to let developers easily achieve durability without having to make complex infrastructure decisions.

The SDK follows three principles:

• No specific infrastructure requirements
• Opt-in durability
• Familiar JavaScript patterns

Through the "use workflow" and "use step" directives, it provides automatic retries, state persistence, and observability without requiring code changes. In a plain Node.js environment, it acts as a no-op with no effect, while it guarantees robust durability only when run in a Workflow DevKit environment.

This means the cost of already-completed steps is never wasted, and long-running tasks can be performed reliably without being bound by serverless function timeout limits. It also provides low-level units of work called primitives, such as fetchTranscriptForAsset, allowing developers to easily incorporate them as individual steps into their own workflows.