From bdbf02a0c4ef3b9b9f9c1f335af9f2b2108f8e74 Mon Sep 17 00:00:00 2001
From: "Aryan Singh K." <70511529+aryansk@users.noreply.github.com>
Date: Mon, 24 Aug 2026 22:52:12 +0530
Subject: [PATCH 1/2] perf: reduce live Markdown re-render churn while
 streaming

Fixes #49

RichUI re-rendered streaming Markdown every 80 new characters under
rich.live.Live, causing hundreds of full-screen re-renders for long
responses (CPU-heavy, visible lag, brutal for vhs demo).

Increase _TRUNCATE_STEP from 80 to 160 and throttle live updates to
max ~10 fps (0.10s min interval via time.monotonic). Keeps partial
output visible while streaming but halves re-renders and caps FPS.

Validation: py_compile passes, git diff --check clean; manual check
shows long response streams with ~50% fewer Live updates and no visual
regression in partial-token display.
---
 gcode/ui.py | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/gcode/ui.py b/gcode/ui.py
index ab3e783..6c970d7 100644
--- a/gcode/ui.py
+++ b/gcode/ui.py
@@ -5,6 +5,8 @@
 display, a permission gate, and status/error output.
 """
 
+import time
+
 import questionary
 from prompt_toolkit import PromptSession
 from prompt_toolkit.formatted_text import HTML
@@ -16,7 +18,8 @@
 from rich.spinner import Spinner
 from rich.text import Text
 
-_TRUNCATE_STEP = 80  # re-render Markdown only after this many new characters
+_TRUNCATE_STEP = 160  # re-render Markdown only after this many new characters
+_TRUNCATE_MIN_INTERVAL = 0.10  # seconds between live updates (max ~10 fps)
 
 # Slash commands available in the interactive menu
 _SLASH_COMMANDS = [
@@ -148,6 +151,7 @@ def prompt(self) -> str:
     def assistant_start(self) -> None:
         self._buffer = ""
         self._last_len = 0
+        self._last_update = 0.0
         self._live = Live(
             Spinner("dots", text="Thinking…"),
             console=self.console,
@@ -160,9 +164,11 @@ def token(self, text: str) -> None:
         if self._live is None:
             return
         self._buffer += text
-        if len(self._buffer) - self._last_len >= _TRUNCATE_STEP:
+        now = time.monotonic()
+        if len(self._buffer) - self._last_len >= _TRUNCATE_STEP and now - self._last_update >= _TRUNCATE_MIN_INTERVAL:
             self._live.update(Markdown(self._buffer))
             self._last_len = len(self._buffer)
+            self._last_update = now
 
     def assistant_end(self) -> None:
         if self._live is None:

From e652e7c44b22d811408ba46946df90433668853c Mon Sep 17 00:00:00 2001
From: shauryagangrade <288927048+shauryagangrade@users.noreply.github.com>
Date: Sat, 29 Aug 2026 22:19:29 +0530
Subject: [PATCH 2/2] perf: reduce live Markdown re-render churn while
 streaming: format + make interval injectable for tests

---
 gcode/ui.py      |  5 ++++-
 tests/test_ui.py | 30 ++++++++++++++++++++++++++++--
 2 files changed, 32 insertions(+), 3 deletions(-)

diff --git a/gcode/ui.py b/gcode/ui.py
index 6c970d7..01d9c47 100644
--- a/gcode/ui.py
+++ b/gcode/ui.py
@@ -165,7 +165,10 @@ def token(self, text: str) -> None:
             return
         self._buffer += text
         now = time.monotonic()
-        if len(self._buffer) - self._last_len >= _TRUNCATE_STEP and now - self._last_update >= _TRUNCATE_MIN_INTERVAL:
+        if (
+            len(self._buffer) - self._last_len >= _TRUNCATE_STEP
+            and now - self._last_update >= _TRUNCATE_MIN_INTERVAL
+        ):
             self._live.update(Markdown(self._buffer))
             self._last_len = len(self._buffer)
             self._last_update = now
diff --git a/tests/test_ui.py b/tests/test_ui.py
index 46e3bfc..36ff5c7 100644
--- a/tests/test_ui.py
+++ b/tests/test_ui.py
@@ -133,16 +133,17 @@ def test_token_renders_markdown_after_threshold(monkeypatch):
 
     fake = _FakeLive()
     monkeypatch.setattr(ui_module, "Live", lambda *args, **kwargs: fake)
+    monkeypatch.setattr(ui_module, "_TRUNCATE_MIN_INTERVAL", 0.0)
 
     ui = RichUI()
     ui.assistant_start()
 
-    # Below the 80-char re-render threshold: no update yet.
+    # Below the 160-char re-render threshold: no update yet.
     ui.token("x" * 40)
     assert fake.updates == []
 
     # Crossing the threshold triggers one Markdown update.
-    ui.token("y" * 50)  # total 90 >= 80
+    ui.token("y" * 150)  # total 190 >= 160
     assert len(fake.updates) == 1
 
     # Ending the stream renders the final text and stops the live region.
@@ -151,6 +152,31 @@ def test_token_renders_markdown_after_threshold(monkeypatch):
     assert len(fake.updates) == 2
 
 
+def test_token_throttled_by_time_interval(monkeypatch):
+    from gcode import ui as ui_module
+
+    fake = _FakeLive()
+    monkeypatch.setattr(ui_module, "Live", lambda *args, **kwargs: fake)
+    monkeypatch.setattr(ui_module, "_TRUNCATE_STEP", 1)
+    monkeypatch.setattr(ui_module, "_TRUNCATE_MIN_INTERVAL", 60.0)
+
+    clock = iter([10.0, 10.05, 70.0])
+    monkeypatch.setattr(ui_module.time, "monotonic", lambda: next(clock))
+
+    ui = RichUI()
+    ui.assistant_start()
+
+    ui.token("a")  # 1 char but only 10s elapsed since 0.0 -> throttled
+    ui.token("b")  # still < 60s elapsed -> throttled
+    assert len(fake.updates) == 0
+
+    ui.token("c")  # 70s elapsed -> update fires
+    assert len(fake.updates) == 1
+
+    ui.assistant_end()
+    assert fake.stopped
+
+
 # -- tool display ----------------------------------------------------------