From fd93bd3414a1725219e14561bc5640b62f9fd4a1 Mon Sep 17 00:00:00 2001
From: Timothy Jaeryang Baek <tim@openwebui.com>
Date: Tue, 14 Apr 2026 11:03:36 -0500
Subject: [PATCH] refac

---
 src/lib/components/notes/NoteEditor/Chat.svelte | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/src/lib/components/notes/NoteEditor/Chat.svelte b/src/lib/components/notes/NoteEditor/Chat.svelte
index 2430f602e1..f8e4b0c914 100644
--- a/src/lib/components/notes/NoteEditor/Chat.svelte
+++ b/src/lib/components/notes/NoteEditor/Chat.svelte
@@ -169,13 +169,20 @@ Based on the user's instruction, update and enhance the existing notes or select
 				: '') +
 			(selectedContent ? `\n<selection>${selectedContent?.text}</selection>` : '');
 
+		// Filter out empty assistant placeholder messages to avoid sending
+		// an empty trailing assistant message as "response prefill", which is
+		// incompatible with enable_thinking in llama.cpp and similar backends.
+		const filteredMessages = messages.filter(
+			(m) => !(m.role === 'assistant' && m.content === '')
+		);
+
 		const chatMessages = JSON.parse(
 			JSON.stringify([
 				{
 					role: 'system',
 					content: `${system}`
 				},
-				...messages
+				...filteredMessages
 			])
 		);