Nice! Thank you for playing around with it.

The chunks are basically words, or chunks of words, but they can be larger than that. E.g. the first chunk is your startWith text if you specified that, and then each subsequent chunk is generally a little piece of text - corresponding to the chunks that are being appended to the output element several times per second.

The render function is specifically for transforming the output into some different form. Whatever you return from that function is what gets displayed - like in this example where we ask the AI for asterisks around actions (since that would be easy for it to generate) but then “render” that text so that the asterisked parts are italicized via HTML. Getting the AI itself to generate HTML is okay, but it has been trained mostly on text, rather than HTML, so it’s probably better to get it to use a “syntax” that it’s more accustomed to, and then we handle the transformation to HTML ourselves with render.

onChunk doesn’t have any effect on the display of the output unless you specifically write some code to do that. It just allows you to run whatever custom code you want every time a new chunk is received.

But yeah you can definitely just use onChunk if you want to manage the “rendering” yourself (e.g. onChunk(data) => outputEl.innerHTML = data.fullTextSoFar.replace(...)), or if you don’t want to change what is displayed, but instead want to do something else for every chunk.

Thanks for the question! I’ve just updated the plugin page with some details on most of the options that are currently available.