brauliogusmao stops the chat from failing silently
When the AI errored mid-reply, users just saw the spinner vanish. This fork makes the system tell them what actually went wrong.
Before this change, if the underlying AI provider hit a problem mid-stream - an expired API key, a rate limit, a context that was too long - the assistant simply stopped typing. No message, no warning, nothing to act on. brauliogusmao's fork now digs the real error message out of the provider's response and pipes it back into the chat bubble where the reply would have been.
A follow-up commit goes further: instead of just passing the raw text through, the backend classifies the error into a small set of categories - out of credits, bad key, rate-limited, context too long, timed out, overloaded - and the frontend shows a clean explanation for each. There's a fair-warning caveat here: the classification relies on matching words in the provider's error text, which can change without notice.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?