diff --git a/index.ts b/index.ts
index f7068c9..93053e8 100644
--- a/index.ts
+++ b/index.ts
@@ -1,9 +1,15 @@
 import os from "node:os";
 import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
-import { streamText } from "ai";
+import { type ModelMessage, streamText } from "ai";
 
 import { renderStream } from "./src/render";
 
+const home = os.homedir();
+const convoFile = Bun.file(`${home}/.aicl_convo.json`);
+const reasoningFile = Bun.file(`${home}/.aicl_reasoning.md`);
+const responsePath = `${home}/.aicl_response.md`;
+const responseFile = Bun.file(responsePath);
+
 const provider = createOpenAICompatible({
   name: "llama.cpp",
   apiKey: "local",
@@ -11,22 +17,31 @@ const provider = createOpenAICompatible({
   includeUsage: true, // Include usage information in streaming responses
 });
 
+const messages: ModelMessage[] = [];
+
+try {
+  const convo = await convoFile.json();
+  messages.push(...convo);
+} catch {}
+
+const prompt = Bun.argv[2] ?? "Introduce yourself.";
+messages.push({ role: "user", content: prompt });
+
 const { fullStream, reasoningText, text } = streamText({
   model: provider("qwen3.5-35b-a3b"),
-  prompt: Bun.argv[2] ?? "Introduce yourself.",
+  messages,
 });
 
 renderStream(fullStream);
 
-const home = os.homedir();
-const reasoningPath = `${home}/.aicl_reasoning.md`;
-const responsePath = `${home}/.aicl_response.md`;
-
 const reasoning = await reasoningText;
-await Bun.file(reasoningPath).write(reasoning ?? "");
+await reasoningFile.write(reasoning ?? "");
 
 const response = await text;
-await Bun.file(responsePath).write(response);
+await responseFile.write(response);
+
+messages.push({ role: "assistant", content: response });
+await convoFile.write(JSON.stringify(messages));
 
 const [numColumns] = process.stdout.getWindowSize();
 Bun.spawnSync({