feat(AI): add done channel for faster response

assistant.go

@@ -134,7 +134,10 @@ func assistantStreamedResponse(request openai.ChatRequest, cb assistantStreamedR
 					if minWaitSatisfied {
 						break Drain
 					}
-					<-minWait
+					select {
+					case <-minWait:
+					case <-resp.Done:
+					}
 					minWaitSatisfied = true
 				}
 			}
@@ -155,7 +158,7 @@ func assistantStreamedResponse(request openai.ChatRequest, cb assistantStreamedR
 			minWaitDurSecs := lo.Min([]int{nUpdates, 4}) + nErrs*3
 			minWait = time.After(time.Duration(minWaitDurSecs) * time.Second)
 
-			// send the partial message
+			// Send the partial message
 			respText := respBuilder.String() + assistantWritingSign
 			if err := cb(respText, false); err != nil {
 				logger.Warnw("failed to send partial update", "error", err)

openai/chat.go

@@ -51,6 +51,7 @@ type ChatResponseStream struct {
 	Created int
 	Model   string
 	Stream  chan string
+	Done    chan struct{}
 	Err     error
 }
 

openai/client.go

@@ -77,11 +77,15 @@ func (c *Client) ChatCompletionStream(request ChatRequest) (*ChatResponseStream,
 		return nil, errors.Errorf("status code: %d, body: %q", resp.StatusCode(), respBodyStr)
 	}
 
-	ret := &ChatResponseStream{Stream: make(chan string, 1024)}
+	ret := &ChatResponseStream{
+		Stream: make(chan string, 1024),
+		Done:   make(chan struct{}),
+	}
 	go func() {
 		defer func() {
 			rbody.Close()
 			close(ret.Stream)
+			close(ret.Done)
 		}()
 
 		var contentBegan bool