Expose version/build metadata and improve provider error messaging

internal/chat/service.go

@@ -37,6 +37,7 @@ type Service struct {
 	model         string
 	temperature   float64
 	stream        bool
+	streamNotice  string
 	history       []openai.ChatMessage
 	sessionID     int64
 	summarySet    bool
@@ -92,6 +93,8 @@ func (s *Service) Send(ctx context.Context, input string, streamHandler openai.C
 	messages := append([]openai.ChatMessage(nil), s.history...)
 	temperature := s.temperature
 
+	s.streamNotice = ""
+
 	req := openai.ChatCompletionRequest{
 		Model:       s.model,
 		Messages:    messages,
@@ -105,11 +108,17 @@ func (s *Service) Send(ctx context.Context, input string, streamHandler openai.C
 	var err error
 	if s.stream {
 		resp, err = s.client.StreamChatCompletion(ctx, req, streamHandler)
+		if err != nil {
+			resp, err = s.handleStreamingFailure(ctx, req, err)
+			if err != nil {
+				return "", s.translateProviderError(err)
+			}
+		}
 	} else {
 		resp, err = s.client.CreateChatCompletion(ctx, req)
-	}
-	if err != nil {
-		return "", err
+		if err != nil {
+			return "", s.translateProviderError(err)
+		}
 	}
 	if len(resp.Choices) == 0 {
 		return "", errors.New("no choices returned from completion")
@@ -135,6 +144,16 @@ func (s *Service) History() []openai.ChatMessage {
 	return historyCopy
 }
 
+// ConsumeStreamingNotice returns any pending streaming notice and clears it.
+func (s *Service) ConsumeStreamingNotice() string {
+	if s == nil {
+		return ""
+	}
+	notice := s.streamNotice
+	s.streamNotice = ""
+	return notice
+}
+
 // StreamingEnabled reports whether streaming completions are configured for this service.
 func (s *Service) StreamingEnabled() bool {
 	if s == nil {
@@ -143,6 +162,74 @@ func (s *Service) StreamingEnabled() bool {
 	return s.stream
 }
 
+func (s *Service) translateProviderError(err error) error {
+	var reqErr *openai.RequestError
+	if !errors.As(err, &reqErr) {
+		return err
+	}
+
+	if guidance, ok := providerStatusGuidance(reqErr.StatusCode()); ok {
+		return errors.New(guidance)
+	}
+
+	return err
+}
+
+func (s *Service) handleStreamingFailure(ctx context.Context, req openai.ChatCompletionRequest, streamErr error) (*openai.ChatCompletionResponse, error) {
+	if s == nil {
+		return nil, streamErr
+	}
+
+	var reqErr *openai.RequestError
+	if !errors.As(streamErr, &reqErr) {
+		return nil, streamErr
+	}
+
+	status := reqErr.StatusCode()
+	if status < 400 || status >= 500 {
+		return nil, streamErr
+	}
+
+	guidance, hasGuidance := providerStatusGuidance(status)
+	message := guidance
+	if !hasGuidance {
+		message = strings.TrimSpace(reqErr.Message())
+		if message == "" {
+			message = strings.TrimSpace(streamErr.Error())
+		}
+		if message == "" {
+			message = "Streaming is unavailable"
+		}
+	}
+	message = fmt.Sprintf("%s\nStreaming has been disabled; responses will be fully buffered.", message)
+
+	s.logger.WarnContext(ctx, "streaming disabled", "status", status, "error", strings.TrimSpace(reqErr.Message()))
+	s.stream = false
+	s.streamNotice = message
+	req.Stream = false
+
+	resp, err := s.client.CreateChatCompletion(ctx, req)
+	if err != nil {
+		return nil, s.translateProviderError(err)
+	}
+	return resp, nil
+}
+
+func providerStatusGuidance(status int) (string, bool) {
+	switch status {
+	case 401:
+		return "Incorrect API key provided.\nVerify API key, clear browser cache, or generate a new key.", true
+	case 429:
+		return "Rate limit reached.\nPace requests and implement exponential backoff.", true
+	case 500:
+		return "Server error.\nRetry after a brief wait; contact support if persistent.", true
+	case 503:
+		return "Engine overloaded.\nRetry request after a brief wait; contact support if persistent.", true
+	default:
+		return "", false
+	}
+}
+
 // Reset clears the in-memory conversation history.
 func (s *Service) Reset() {
 	if s == nil {