diff --git a/public/src/chat.js b/public/src/chat.js
index dfd869e..8304380 100644
--- a/public/src/chat.js
+++ b/public/src/chat.js
@@ -33,7 +33,7 @@ function calculateAvailableTokens(chatHistory, maxTokens) {
 	console.log(`Token calculation: max=${maxTokens}, input≈${inputTokens}, buffer=${buffer}, available≈${availableTokens}`);
 	return availableTokens;
 }
-import { initializeChatHistory, saveConversationHistory } from "./storage.js";
+import { initializeChatHistory, saveConversationHistory, getCustomPromptEcho } from "./storage.js";
 import { generateTitleForTTS, speakText } from "./tts.js";
 
 // Helper function to add copy buttons to code blocks
@@ -106,19 +106,25 @@ export async function* sendMessage(message) {
 		model = getSelectedModel();
 	}
 
+	// Build messages for LLM: chatHistory + echo (echo is only for LLM attention, not stored)
+	const echo = getCustomPromptEcho();
+	const messagesForLLM = echo
+		? [...chatHistory, { role: "assistant", content: echo }]
+		: chatHistory;
+
 	const modelMaxTokens = getSelectedModelMaxTokens();
-	const calculatedAvailable = calculateAvailableTokens(chatHistory, modelMaxTokens);
-	
+	const calculatedAvailable = calculateAvailableTokens(messagesForLLM, modelMaxTokens);
+
 	// Cap to model's max completion tokens limit
 	const maxCompletionTokens = getSelectedModelMaxCompletionTokens();
 	// Use the full calculated available space or the model's limit, whichever is smaller
 	const finalAvailable = Math.min(calculatedAvailable, maxCompletionTokens);
-	
+
 	console.log("Chat using tokens:", finalAvailable, "calculated:", calculatedAvailable, "max completion:", maxCompletionTokens);
-	
+
 	const requestBody = JSON.stringify({
 		model: model,
-		messages: chatHistory,
+		messages: messagesForLLM,
 		temperature: 0.5,
 		max_tokens: finalAvailable,
 		stream: true,
@@ -273,20 +279,26 @@ export async function* sendMessageWithCustomHistory(customHistory) {
 		model = getSelectedModel();
 	}
 
+	// Build messages for LLM: customHistory + echo (echo is only for LLM attention, not stored)
+	const echo = getCustomPromptEcho();
+	const messagesForLLM = echo
+		? [...customHistory, { role: "assistant", content: echo }]
+		: customHistory;
+
 	const modelMaxTokens = getSelectedModelMaxTokens();
-	const calculatedAvailable = calculateAvailableTokens(customHistory, modelMaxTokens);
-	
+	const calculatedAvailable = calculateAvailableTokens(messagesForLLM, modelMaxTokens);
+
 	// Cap to model's max completion tokens limit
 	const maxCompletionTokens = getSelectedModelMaxCompletionTokens();
 	const finalAvailable = Math.min(calculatedAvailable, maxCompletionTokens);
 	console.log("Chat (custom history) using tokens:", finalAvailable, "calculated:", calculatedAvailable, "max completion:", maxCompletionTokens);
-	
+
 	const response = await fetch(apiUrl, {
 		method: "POST",
 		headers: headers,
 		body: JSON.stringify({
 			model: model,
-			messages: customHistory,
+			messages: messagesForLLM,
 			max_tokens: finalAvailable,
 			stream: true,
 		}),
diff --git a/public/src/storage.js b/public/src/storage.js
index f51f089..57046ca 100644
--- a/public/src/storage.js
+++ b/public/src/storage.js
@@ -35,7 +35,8 @@ export function loadConversationHistory() {
 }
 
 // Get just the custom prompt for attention echo (not full system message with page content)
-function getCustomPromptEcho() {
+// This is exported so chat.js can add it only to LLM requests, not stored history
+export function getCustomPromptEcho() {
 	if (typeof window !== "undefined" && window.UNCLOSEAI_SYSTEM_PROMPT) {
 		return window.UNCLOSEAI_SYSTEM_PROMPT;
 	}
@@ -45,20 +46,16 @@ function getCustomPromptEcho() {
 export async function clearConversationHistory() {
 	localStorage.removeItem(getPageSpecificKey("hermes-conversation-history"));
 	const systemContent = await getSystemMessage();
-	const echo = getCustomPromptEcho();
-	const history = [{ role: "system", content: systemContent }];
-	if (echo) history.push({ role: "assistant", content: echo });
-	return history;
+	// Echo is NOT added here - it's only added at LLM request time in chat.js
+	return [{ role: "system", content: systemContent }];
 }
 
 // Initialize chat history
 export async function initializeChatHistory() {
 	const systemContent = await getSystemMessage();
-	const echo = getCustomPromptEcho();
-	const history = [
+	// Echo is NOT added here - it's only added at LLM request time in chat.js
+	return [
 		{ role: "system", content: systemContent },
 		...loadConversationHistory(),
 	];
-	if (echo) history.push({ role: "assistant", content: echo });
-	return history;
 }
diff --git a/public/src/uncloseai-embed-modal.js b/public/src/uncloseai-embed-modal.js
index 690d2f0..b1124cd 100644
--- a/public/src/uncloseai-embed-modal.js
+++ b/public/src/uncloseai-embed-modal.js
@@ -1496,12 +1496,11 @@ You have complete knowledge of this page content and can reference any details,
 				const { getSystemMessage } = await import("./config.js");
 
 				// Create new history with updated system message and loaded conversation
+				// Echo is NOT added here - it's only added at LLM request time in chat.js
 				const systemContent = await getSystemMessage();
-				const echo = (typeof window !== "undefined" && window.UNCLOSEAI_SYSTEM_PROMPT) ? window.UNCLOSEAI_SYSTEM_PROMPT : null;
 				const newHistory = [
 					{ role: "system", content: systemContent },
 					...history,
-					...(echo ? [{ role: "assistant", content: echo }] : []),
 				];
 
 				// Update the chat.js module's history
@@ -1644,14 +1643,13 @@ You have complete knowledge of this page content and can reference any details,
 								await saveConversationHistory(currentHistory);
 
 								// Update the chat.js module history to match localStorage
+								// Echo is NOT added here - it's only added at LLM request time in chat.js
 								const { updateChatHistory } = await import("./chat.js");
 								const { getSystemMessage } = await import("./config.js");
 								const systemContent = await getSystemMessage();
-								const echo = (typeof window !== "undefined" && window.UNCLOSEAI_SYSTEM_PROMPT) ? window.UNCLOSEAI_SYSTEM_PROMPT : null;
 								const newHistory = [
 									{ role: "system", content: systemContent },
 									...currentHistory,
-									...(echo ? [{ role: "assistant", content: echo }] : []),
 								];
 								await updateChatHistory(newHistory);
 							}
@@ -1965,13 +1963,12 @@ You have complete knowledge of this page content and can reference any details,
 			const { updateChatHistory, getChatHistory } = await import("./chat.js");
 			const { getSystemMessage } = await import("./config.js");
 
-			// Create new history with system message, intro, and echo for attention
+			// Create new history with system message and intro
+			// Echo is NOT added here - it's only added at LLM request time in chat.js
 			const systemContent = await getSystemMessage();
-			const echo = (typeof window !== "undefined" && window.UNCLOSEAI_SYSTEM_PROMPT) ? window.UNCLOSEAI_SYSTEM_PROMPT : null;
 			const newHistory = [
 				{ role: "system", content: systemContent },
 				{ role: "assistant", content: response },
-				...(echo ? [{ role: "assistant", content: echo }] : []),
 			];
 
 			// Update the chat.js module's history