Works well and seems more polished than other offerings. There's some small bugs i ran into that I'm not sure why - for ex. some specific questions will prompt a message saying you've hit your quota or something (i haven't) but asking something different will be ok. Sometimes mid-sentence it'll saying something went wrong and contact ChatNode's email. But these bugs only really seemed to happen when I used my own key for some reason. Anyways, those are minor and overall it's good but I have a few suggestions:

- Ability to change the base prompt ChatNode uses, mainly because I kinda don't like how it keeps asking 'how may i assist you' after every message - just want to give it an answer avoid sounding so repetitive. If you tell it not to do that in your own prompt it'll still do it, so there's some kind of master prompt that is limiting customizing the behavior.

- For the API, ability to select which data to include/exclude. Sometimes I just want to have it look at one doc/url or remove some others and ask the question again because some questions require certain context for the answer to be useful.

- More info in the docs on how it searches for answers - is it based on embeddings for each paragraph and a vector search? And how best to structure the data. It seems to work best for FAQ type data where a question+answer is in a specific place. It'll be nice to know it works so we know how best to structure the data.

- Ability to add an external API as a data source and any other advanced controls (chaining, preprocessing, etc) would be much appreciated and differentiate from all the other similiar services