Merge pull request #933 from harshalmore31/feat/921-stream

Improved the formatting of streaming output and fixed the print_on functionality

stream_example.py

@@ -6,7 +6,7 @@ agent = Agent(
     model_name="gpt-4o-mini",
     streaming_on=True,  # 🔥 This enables real streaming!
     max_loops=1,
-    print_on=False,  # By Default its False, raw streaming !!
+    print_on=False,
     output_type="all",
 )
 

swarms/structs/agent.py

@@ -2518,13 +2518,8 @@ class Agent:
                 ) and not isinstance(streaming_response, str):
                     # Check print_on parameter for different streaming behaviors
                     if self.print_on is False:
-                        # Show raw streaming text without formatting panels
+                        # Silent streaming - no printing, just collect chunks
                         chunks = []
-                        print(
-                            f"\n{self.agent_name}: ",
-                            end="",
-                            flush=True,
-                        )
                         for chunk in streaming_response:
                             if (
                                 hasattr(chunk, "choices")
@@ -2533,11 +2528,7 @@ class Agent:
                                 content = chunk.choices[
                                     0
                                 ].delta.content
-                                print(
-                                    content, end="", flush=True
-                                )  # Print raw streaming text
                                 chunks.append(content)
-                        print()  # New line after streaming completes
                         complete_response = "".join(chunks)
                     else:
                         # Collect chunks for conversation saving
@@ -2557,7 +2548,7 @@ class Agent:
                         complete_response = formatter.print_streaming_panel(
                             streaming_response,
                             title=f"🤖 Agent: {self.agent_name} Loops: {current_loop}",
-                            style="bold cyan",
+                            style=None,  # Use random color like non-streaming approach
                             collect_chunks=True,
                             on_chunk_callback=on_chunk_received,
                         )
@@ -2796,19 +2787,20 @@ class Agent:
         return self.role
 
     def pretty_print(self, response: str, loop_count: int):
+        if self.streaming_on is True:
+            # Skip printing here since real streaming is handled in call_llm
+            # This avoids double printing when streaming_on=True
+            return
+        
         if self.print_on is False:
-            if self.streaming_on is True:
-                # Skip printing here since real streaming is handled in call_llm
-                # This avoids double printing when streaming_on=True
-                pass
-            elif self.no_print is True:
-                pass
-            else:
-                # logger.info(f"Response: {response}")
-                formatter.print_panel(
-                    f"{self.agent_name}: {response}",
-                    f"Agent Name {self.agent_name} [Max Loops: {loop_count} ]",
-                )
+            # Silent mode - no printing at all
+            return
+        else:
+            # Use formatted panels (default behavior when print_on=True)
+            formatter.print_panel(
+                f"{self.agent_name}: {response}",
+                f"Agent Name {self.agent_name} [Max Loops: {loop_count} ]",
+            )
 
     def parse_llm_output(self, response: Any):
         """Parse and standardize the output from the LLM.

swarms/utils/formatter.py

@@ -156,7 +156,7 @@ class Formatter:
         self,
         streaming_response,
         title: str = "🤖 Agent Streaming Response",
-        style: str = choose_random_color(),
+        style: str = None,
         collect_chunks: bool = False,
         on_chunk_callback: Optional[Callable] = None,
     ) -> str:
@@ -167,17 +167,21 @@ class Formatter:
         Args:
             streaming_response: The streaming response generator from LiteLLM.
             title (str): Title of the panel.
-            style (str): Style for the panel border.
+            style (str): Style for the panel border (if None, will use random color).
             collect_chunks (bool): Whether to collect individual chunks for conversation saving.
             on_chunk_callback (Optional[Callable]): Callback function to call for each chunk.
 
         Returns:
             str: The complete accumulated response text.
         """
+        # Get random color similar to non-streaming approach
+        random_color = choose_random_color()
+        panel_style = f"bold {random_color}" if style is None else style
+        text_style = random_color
 
         def create_streaming_panel(text_obj, is_complete=False):
             """Create panel with proper text wrapping using Rich's built-in capabilities"""
-            panel_title = f"[bold cyan]{title}[/bold cyan]"
+            panel_title = f"[white]{title}[/white]"
             if is_complete:
                 panel_title += " [bold green]✅[/bold green]"
 
@@ -190,7 +194,7 @@ class Formatter:
             panel = Panel(
                 display_text,
                 title=panel_title,
-                border_style=style,
+                border_style=panel_style,
                 padding=(1, 2),
                 width=self.console.size.width,  # Rich handles wrapping automatically
             )
@@ -214,9 +218,9 @@ class Formatter:
                         and part.choices
                         and part.choices[0].delta.content
                     ):
-                        # Add ONLY the new chunk to the Text object
+                        # Add ONLY the new chunk to the Text object with random color style
                         chunk = part.choices[0].delta.content
-                        streaming_text.append(chunk, style="white")
+                        streaming_text.append(chunk, style=text_style)
                         complete_response += chunk
 
                         # Collect chunks if requested