Getting a connection working today requires a lot of trial and error with minimal feedback. Several targeted improvements would significantly reduce time-to-resolution:

Richer Form Guidance: Field-level documentation or hints for provider-specific requirements (e.g., Azure OpenAI's deployment name format, API version strings). These requirements aren't obvious and differ across providers.

Detailed Error Messages: When a connection test fails, surface the actual failure reason rather than a generic error. Knowing whether a failure is an auth issue, a bad endpoint, or a malformed request saves significant diagnostic time.

Outgoing Request URI Visibility: Show the full URI being constructed so users can verify it matches what the provider expects. This is especially useful for Azure OpenAI where the endpoint format is highly specific.

Raw Response Viewer: The ability to see the raw HTTP response from the provider (status code, headers, body) would allow users to self-diagnose without needing to involve support or set up external tooling to intercept traffic.