diff --git a/discord/bot.ts b/discord/bot.ts
index 70f8f40..1a981f6 100644
--- a/discord/bot.ts
+++ b/discord/bot.ts
@@ -214,23 +214,45 @@ async function requestLLMResponse(messages)
         queryParams.append(field, config["llmconf"].llmSettings[field]);
     }
     const llmEndpoint = `${process.env.LLM_HOST}/?${queryParams.toString()}`;
-    const messageList = messages.map((m: Message) => {
+    let messageList = await Promise.all(
+        messages.map(async (m: Message) => {
 	    let role = 'user';
 	    if (m.author.id === process.env.CLIENT) {
 		    role = 'assistant';
 	    } else if (m.author.bot) {
 		    return null;
-	    } else if (KNOWN_USERNAMES.includes(m.author.username)) {
-		    role = m.author.username;
+	    /* } else if (KNOWN_USERNAMES.includes(m.author.username)) {
+	           role = m.author.username; */
 	    }
-	    return { role, content: m.cleanContent };
-    });
+	    // fetch replied-to message, if there is one, and prompt it as such
+		let cleanContent = m.cleanContent;
+		if (m.type == MessageType.Reply && m.reference) {
+			// what about deeply nested replies? could possibly be recursive?
+			const repliedToMsg = await m.fetchReference();
+			if (repliedToMsg) {
+				const repliedToMsgLines = repliedToMsg.cleanContent.split('\n');
+				cleanContent = `> ${repliedToMsgLines.join('\n> ')}\n${cleanContent}`;
+			}
+		}
+
+	    return { role, content: cleanContent };
+        })
+    );
+    messageList = messageList.filter(x => !!x);
+    
+    // at the beginning, inject the system prompt
+    // at the end, start our text generation as a reply to the most recent msg from history
+    const replyContext = `> ${messageList[messageList.length - 1].content.split('\n').join('\n> ')}\n`;
     const reqBody = [
         {
             "role": "system",
             "content": config["llmconf"].sys_prompt
         },
-        ...messageList.filter(x => x)
+        ...messageList,
+        {
+            "role": "assistant",
+            "content": replyContext
+        }
     ];
     logInfo("[bot] Requesting LLM response with message list: " + reqBody.map(m => m.content));
     const res = await fetch(llmEndpoint, {