diff --git a/examples/cli/src/index.ts b/examples/cli/src/index.ts
index 45008b6..10a69ba 100644
--- a/examples/cli/src/index.ts
+++ b/examples/cli/src/index.ts
@@ -6,32 +6,52 @@ import { createOpenAiClient } from './app';
 
 export type Argv = Record<string, unknown>;
 
-const receiveData = (aiClient: PlatformEventEmitter, theContent: string, argv: Record<string, unknown>) => new Promise<void>((r2, rj) => {
+const receiveData = (
+  aiClient: PlatformEventEmitter,
+  theContent: string,
+  argv: Record<string, unknown>,
+  memory: OpenAi.MessageObject[],
+) => new Promise<void>((r2, rj) => {
   let completionTokens: number;
 
+  const normalizedChatMessage = OpenAi.normalizeChatMessage(theContent);
+
   const model = (
     argv.model as OpenAi.ChatCompletionModel
     ?? OpenAi.ChatCompletionModel.GPT_3_5_TURBO
   );
 
+  let assistantMessage: Partial<OpenAi.MessageObject>;
   aiClient.on<OpenAi.ChatCompletionChunkDataEvent>('data', (d) => {
     if (d.choices?.[0]?.delta?.role) {
+      assistantMessage.role = d.choices[0].delta.role;
+      assistantMessage.content = '';
       process.stdout.write(`${d.choices[0].delta.role}:\n`);
     }
 
     if (d.choices?.[0]?.delta?.content) {
       completionTokens += 1;
+      assistantMessage.content += d.choices[0].delta.content;
       process.stdout.write(d.choices[0].delta.content);
     }
   });
 
   aiClient.on('end', () => {
+    normalizedChatMessage.forEach((m) => {
+      memory.push(m);
+    });
+    memory.push(assistantMessage as OpenAi.MessageObject);
     process.stdout.write('\n\n');
     const { length: promptTokens } = OpenAi.getPromptTokens(
-      OpenAi.normalizeChatMessage(theContent),
+      normalizedChatMessage,
       model,
     );
-    process.stdout.write(`info:\n${promptTokens} prompt tokens\n${completionTokens} completion tokens\n\n`);
+    process.stdout.write(`info:\n${promptTokens} prompt tokens\n${completionTokens} completion tokens`);
+    process.stdout.write(`info:\n${completionTokens} completion tokens`);
+    if (argv.memory) {
+      process.stdout.write(`\n${memory.length} memory items`);
+    }
+    process.stdout.write('\n\n');
     // TODO count tokens
     r2();
   });
@@ -43,9 +63,24 @@ const receiveData = (aiClient: PlatformEventEmitter, theContent: string, argv: R
   });
 
   completionTokens = 0;
+  assistantMessage = {};
+  const effectiveMemory = (
+    argv.memory === true
+      ? memory
+      : memory.slice(-(argv.memory as number))
+  );
   aiClient.createChatCompletion({
     model,
-    messages: theContent,
+    messages: (
+      argv.memory
+        ? [
+          ...effectiveMemory,
+          ...normalizedChatMessage,
+        ]
+        : [
+          ...normalizedChatMessage,
+        ]
+    ),
     temperature: argv.temperature as number,
     maxTokens: argv.maxTokens as number,
     topP: argv.topP as number,
@@ -55,6 +90,7 @@ const receiveData = (aiClient: PlatformEventEmitter, theContent: string, argv: R
 const main = (argv: Argv) => new Promise<number>(async (resolve) => {
   let done = false;
   let resolveResult = 0;
+  const memory = [] as OpenAi.MessageObject[];
 
   process.stdout.write('Welcome to mio-ai CLI!\n\n');
 
@@ -85,14 +121,14 @@ const main = (argv: Argv) => new Promise<number>(async (resolve) => {
         break;
       }
       process.stdout.write('\n');
-      await receiveData(aiClient, content, argv);
+      await receiveData(aiClient, content, argv, memory);
     } catch {
       resolveResult = -1;
       done = true;
     }
   } while (!done);
 
-  process.stdout.write('Bye!\n\n');
+  process.stdout.write('\ninfo:\nUser exited loop\n\n');
   resolve(resolveResult);
 });