import os from "node:os";
import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
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",
  baseURL: "http://127.0.0.1:8080/v1",
  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, totalUsage } = streamText({
  model: provider("qwen3.5-35b-a3b"),
  messages,
});

const start = Date.now();

renderStream(fullStream);

const reasoning = await reasoningText;
const response = await text;

const stop = Date.now();

await reasoningFile.write(reasoning ?? "");
await responseFile.write(response);

messages.push({ role: "assistant", content: response });
await convoFile.write(JSON.stringify(messages));

const [numColumns] = process.stdout.getWindowSize();
Bun.spawnSync({
  cmd: ["glow", "-w", `${numColumns}`, responsePath],
  stdio: ["inherit", "inherit", "inherit"],
});

const { inputTokens, outputTokens } = await totalUsage;
const context = (inputTokens ?? 0) / 1000;
const output = (outputTokens ?? 0) / 1000;
const time = (stop - start) / 1000;
process.stderr.write(
  `Context: ${context.toFixed(1)}k, Output: ${output.toFixed(1)}k, Time: ${time.toFixed(2)}s\n`,
);