stevekrouse/valwriter_react_clientside

streaming
- we can't stream into an actual val - we can only write a full piece of text, i guess we can stream back the code so we don't get bored while we wait
send the code of the valwriter back to gpt
- only if it's related, might need some threads
send errors, logs back to gpt
get screenshots of the output back to gpt
experiment with applying diffs instead of regenerating from scratch every time
- could also have it as a conversation as the main thing and only the diffs get applied or the whole code gets replaced, maybe tool use it the key here... which does make it seem like a custom gpt may be the better fit...