It would be great to see logging or tracing for BrowserGPT to better capture silent fails. Inspired by @BaseInfinity