import sys
from llama_cpp import Llama          # pip install llama-cpp-python
from tools import parse_call, run_tool
llm = Llama(model_path=sys.argv[1], n_ctx=2048, verbose=False)
msgs = []
while True:
    q = input("you> ").strip()
    if not q: break
    msgs.append({"role": "user", "content": q})
    for _ in range(3):
        out = llm.create_chat_completion(messages=msgs, max_tokens=700, temperature=0.6, top_p=0.95)["choices"][0]["message"]["content"]
        if "<think>" not in out: out = "<think>\n" + out      # the chat template already opened the think block
        msgs.append({"role": "assistant", "content": out})
        call = parse_call(out)
        if not call: break
        try: res = run_tool(call["name"], call["arguments"])
        except Exception as e: res = "error: " + str(e)
        msgs.append({"role": "tool", "content": res})
    print(out.split("</think>")[-1].strip())