agent in 50 lines

agent.py

import os
import json
import re
import subprocess
import urllib.request as req
import urllib.error
from urllib.parse import urlencode

# ---------------------------------------------------------------------------
# web.py (inlined)
# ---------------------------------------------------------------------------

UA = {
    "User-Agent": "python-requests/2.31.0"
}

def fetch(url: str, method: str = 'GET', data: dict = None, headers: dict = None) -> str:
    headers = {**UA, **(headers or {})}
    headers = {**(headers or {})}
    if data and method == 'GET':
        url = f"{url}?{urlencode(data)}"
    request = req.Request(url, method=method, headers=headers)
    if data and method != 'GET':
        request.add_header("Content-Type", "application/json")
        request.data = json.dumps(data).encode()
    try:
        return req.urlopen(request).read().decode()
    except urllib.error.HTTPError as e:
        body = e.read().decode(errors="replace")
        raise Exception(
            f"HTTP {e.code} {e.reason}\n"
            f"URL: {e.url}\n"
            f"Response:\n{body}"
        )

def post_json(url: str, data: dict, headers: dict = None) -> dict:
    response = fetch(url, method='POST', data=data, headers=headers)
    return json.loads(response)

# ---------------------------------------------------------------------------
# llm.py (inlined)
# ---------------------------------------------------------------------------

API_URL = "https://openrouter.ai/api/v1/chat/completions"
API_KEY = os.getenv("OPENROUTER_API_KEY")

FLASH = "deepseek/deepseek-v4-flash"
PRO   = "xiaomi/mimo-v2.5-pro"

PROVIDER_MAP = {
    FLASH: "alibaba",
    PRO:   "xiaomi",
}

def chat(
    messages,
    model: str = FLASH,
    reasoning_effort: str = "default",
    **kw,
) -> str:
    def _to_messages(input):
        if isinstance(input, str):
            return [{"role": "user", "content": input}]
        return input

    payload = {
        "model": model,
        "messages": messages,
        "stream": False,
        **kw,
    }

    provider = PROVIDER_MAP.get(model)
    if provider:
        payload["provider"] = {
            "only": [provider],
            "allow_fallbacks": False,
        }

    if reasoning_effort == "default":
        payload["reasoning"] = {
            "enabled": True,
        }
    elif reasoning_effort is not None:
        payload["reasoning"] = {
            "effort": reasoning_effort,
        }

    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {API_KEY}",
    }

    result = post_json(API_URL, payload, headers)
    return result["choices"][0]["message"]["content"]

def chat_flash(messages, **kw):
    return chat(messages, model=FLASH, **kw)

def chat_flash_nr(messages, **kw):
    return chat(messages, model=FLASH, reasoning_effort="none", **kw)

def chat_pro(messages, **kw):
    return chat(messages, model=PRO, **kw)

def chat_pro_nr(messages, **kw):
    return chat(messages, model=PRO, reasoning_effort="none", **kw)

# ---------------------------------------------------------------------------
# main.py logic
# ---------------------------------------------------------------------------

def get_cmd(lm_output: str) -> str:
    matches = re.findall(
        r"```bash-action\s*\n(.*?)\n```",
        lm_output,
        re.DOTALL
    )
    return matches[0].strip() if matches else ""

def run_cmd(command: str) -> str:
    result = subprocess.run(
        command,
        shell=True,
        text=True,
        env=os.environ,
        encoding='utf-8',
        errors='replace',
        stdout=subprocess.PIPE,
        stderr=subprocess.STDOUT,
        timeout=30,
    )
    return result.stdout

def main():
    # Use chat_flash as the default model
    SYSTEM = "You are a helpful assistant. When you want to run a command, wrap it in ```bash-action\n<command>\n```. To finish a turn, run the exit command."

    messages = []
    messages.append({"role": "system", "content": SYSTEM})

    while True:
        user_msg = input("> ")
        messages.append({"role": "user", "content": user_msg})
        print()

        while True:
            lm_output = chat_flash(messages)
            print("Agent:\n", lm_output, "\n")
            messages.append({"role": "assistant", "content": lm_output})
            cmd = get_cmd(lm_output)
            if not cmd:
                break
            if cmd == "exit":
                return
            output = run_cmd(cmd)
            print("Output:\n", output, "\n")
            messages.append({"role": "user", "content": output})

if __name__ == '__main__':
    main()

return chat(messages, model=FLASH, reasoning_effort="none", **kw)

Shouldn’t it be reasoning_effort=None? (in the chat function above you compare with None)