When you are on a slow connection, the new assistant at first returns the error "We did not get a response from the server", but if you wait for the prompt endpoint request to finish, it works, and the AI response overwrites the error message.
You can verify this by using a slow connection and getting the error, then checking in the browser developer tools that the HTTP request is still pending. Waiting, you will eventually get the response, and it will appear, overwriting the error.
Keep in mind I am in a network that blocks WebSockets, so the assistant should gracefully handle an HTTP only fallback.
At first:
Then, after waiting a bit:
Do not display an error if the prompt endpoint request is still pending, as it will eventually get a response on slow connections.
Replace the "Connecting..." status after sending a query, as the assistant is clearly not just connecting to the server, but instead sent a prompt HTTP request and is waiting for the response. An appropriate message should be something like "Waiting for answer...".
Gracefully handle fallback to HTTP if WebSockets is not available.
