From a4c9b069dcebe59fa0a2e08c42349083c803bb75 Mon Sep 17 00:00:00 2001
From: Brooklyn Nicholson <brooklyn.bb.nicholson@gmail.com>
Date: Wed, 15 Jul 2026 14:35:15 -0400
Subject: [PATCH] fix(desktop): stop reasoning text flashing/re-typing on every
 delta
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Reasoning ("Thinking") rendered through MarkdownTextContent with a smooth
typewriter reveal. TextMessagePartProvider mints a fresh part object on every
text change, and useSmooth resets its reveal to empty whenever the part
identity changes — so each reasoning delta restarted the animation from the
first character (h / hell / hello wo ...). Token-streaming reasoners
(R1/Qwen/GLM/Claude thinking) fire a delta per token and flash hard; GPT-5's
coarse reasoning summary updates too rarely to notice, which is why it looked
fine.

Drop smooth on the reasoning path so it plain-appends, exactly like the
assistant answer already does. Removes the now-dead smooth plumbing too.
---
 .../components/assistant-ui/markdown-text.tsx | 23 ++++++++-----------
 1 file changed, 9 insertions(+), 14 deletions(-)

diff --git a/apps/desktop/src/components/assistant-ui/markdown-text.tsx b/apps/desktop/src/components/assistant-ui/markdown-text.tsx
index f9eeae55b63e..8f17685108ff 100644
--- a/apps/desktop/src/components/assistant-ui/markdown-text.tsx
+++ b/apps/desktop/src/components/assistant-ui/markdown-text.tsx
@@ -1,6 +1,6 @@
 'use client'
 
-import { type SmoothOptions, TextMessagePartProvider, useMessagePartText } from '@assistant-ui/react'
+import { TextMessagePartProvider, useMessagePartText } from '@assistant-ui/react'
 import {
   parseMarkdownIntoBlocks,
   type StreamdownTextComponents,
@@ -342,7 +342,6 @@ interface MarkdownTextSurfaceProps {
   containerClassName?: string
   containerProps?: ComponentProps<'div'>
   defer?: boolean
-  smooth?: boolean | SmoothOptions
 }
 
 // Headings shrink to chat scale rather than the prose default (h1≈xl). Kept
@@ -391,7 +390,7 @@ function HugeTextFallback({ containerClassName, text }: { containerClassName?: s
   )
 }
 
-function MarkdownTextSurface({ containerClassName, containerProps, defer, smooth }: MarkdownTextSurfaceProps) {
+function MarkdownTextSurface({ containerClassName, containerProps, defer }: MarkdownTextSurfaceProps) {
   const { status, text } = useMessagePartText()
   const isStreaming = status.type === 'running'
 
@@ -531,29 +530,25 @@ function MarkdownTextSurface({ containerClassName, containerProps, defer, smooth
       parseMarkdownIntoBlocksFn={parseMarkdownIntoBlocksCached}
       plugins={plugins}
       preprocess={preprocessWithTailRepair}
-      smooth={smooth}
     />
   )
 }
 
-const SMOOTH_OPTIONS: SmoothOptions = {
-  drainMs: 500,
-  maxCharsPerFrame: 30,
-  minCommitMs: 33
-}
-
 interface MarkdownTextContentProps extends MarkdownTextSurfaceProps {
   isRunning: boolean
   text: string
 }
 
 export function MarkdownTextContent({ isRunning, text, ...surfaceProps }: MarkdownTextContentProps) {
-  // Same path as the assistant answer. A reasoning-only smoothing wrapper used
-  // to sit here but stalled its char-reveal at empty (the part stays running
-  // the whole message), blanking the Thinking widget.
+  // No `smooth` on purpose — same as the assistant answer. `TextMessagePartProvider`
+  // mints a fresh part object on every `text` change, and useSmooth resets its
+  // reveal to empty whenever the part identity changes, so a smoothed reasoning
+  // stream re-types from the first character on every delta (the flash). Token-
+  // streaming reasoners (R1/Qwen/GLM/Claude thinking) hit it hardest; GPT-5's
+  // coarse summary updates too rarely to notice. Plain append matches the answer.
   return (
     <TextMessagePartProvider isRunning={isRunning} text={text}>
-      <MarkdownTextSurface defer smooth={SMOOTH_OPTIONS} {...surfaceProps} />
+      <MarkdownTextSurface defer {...surfaceProps} />
     </TextMessagePartProvider>
   )
 }