What is hosted-models?

Runway's hosted-models SDK lets you talk to RunwayML's hosted AI models from your JavaScript app with just a few lines of code. If you've deployed a machine learning model on RunwayML and want to send it input from a website or a Node.js backend, this library handles the connection for you. Under the hood, it's a lightweight wrapper around Runway's HTTP API. You create a model object by passing in a URL and, if the model is private, an access token. From there you have four methods: check what inputs and outputs the model expects, send a query with your data, check whether the model is awake, and wait until it's ready. The library also handles some practical concerns like retrying failed requests when the model is still waking up and providing human-readable error messages when something goes wrong. This is aimed at developers building apps or prototypes that use RunwayML models but who don't want to deal with writing raw HTTP requests. For example, if you have a text generation model hosted on RunwayML and want to display the results on a web page, you can include the library via a script tag, pass in your model URL and token, call the query method with a prompt, and display the returned text. It works the same way in a Node.js environment via npm. One thing worth noting is the concept of model sleep states. Hosted models can be asleep, waking up, or awake. A sleeping model takes longer to respond, so the library gives you tools to handle that gracefully, like showing a loading screen while you wait for the model to wake. The library is intentionally minimal with just four methods, which keeps things simple but means you're limited to what those methods cover.