Fix streaming render bugs in chat and query

Two issues in the Live-based markdown streaming:

1. After a tool call, the new Live instance re-rendered the entire
   response so far because `collected` was never reset. Text before
   and after the tool call would be rendered together as one markdown
   block in the new Live region, with no separator between the two
   parts. Track the current segment separately from the full answer.

2. The tool_call_item handler started a new Live immediately after
   printing the tool line, which then had to be stopped in the text
   handler before the blank line. The empty Live start/stop plus the
   explicit `print()` produced two blank lines. Drop the premature
   start and let the next text delta create the new Live.

Author: rejojer

openkb/agent/chat.py

@@ -238,6 +238,7 @@ async def _run_turn(
 
     print()
     collected: list[str] = []
+    segment: list[str] = []
     last_was_text = False
     need_blank_before_text = False
 
@@ -266,18 +267,18 @@ def _start_live() -> Any:
                     text = event.data.delta
                     if text:
                         if need_blank_before_text:
-                            if live:
-                                live.stop()
-                                live = None
+                            if console is not None:
                                 print()
+                                segment = []
                                 live = _start_live()
                             else:
                                 sys.stdout.write("\n")
                             need_blank_before_text = False
                         collected.append(text)
+                        segment.append(text)
                         last_was_text = True
                         if live:
-                            live.update(Markdown("".join(collected), code_theme="monokai"))
+                            live.update(Markdown("".join(segment), code_theme="monokai"))
                         else:
                             sys.stdout.write(text)
                             sys.stdout.flush()
@@ -299,7 +300,6 @@ def _start_live() -> Any:
                         live.stop()
                         live = None
                     _fmt(style, ("class:tool", _format_tool_line(name, args) + "\n"))
-                    live = _start_live()
                     need_blank_before_text = True
     finally:
         if live:

openkb/agent/query.py

@@ -142,15 +142,17 @@ def _start_live() -> Live | None:
 
     result = Runner.run_streamed(agent, question, max_turns=MAX_TURNS)
     collected: list[str] = []
+    segment: list[str] = []
     try:
         async for event in result.stream_events():
             if isinstance(event, RawResponsesStreamEvent):
                 if isinstance(event.data, ResponseTextDeltaEvent):
                     text = event.data.delta
                     if text:
                         collected.append(text)
+                        segment.append(text)
                         if live:
-                            live.update(Markdown("".join(collected), code_theme="monokai"))
+                            live.update(Markdown("".join(segment), code_theme="monokai"))
                         else:
                             sys.stdout.write(text)
                             sys.stdout.flush()
@@ -164,6 +166,7 @@ def _start_live() -> Live | None:
                         live = None
                     sys.stdout.write(f"\n[tool call] {raw.name}({args})\n\n")
                     sys.stdout.flush()
+                    segment = []
                     live = _start_live()
                 elif item.type == "tool_call_output_item":
                     pass