AI agent kaise banaye Python mein: Hindi (Hinglish) guide aur poora code

import json

MODEL = "claude-sonnet-5-5"  # naya model aaye toh ID Anthropic docs se badaliye

# 1) Tools: har tool ek Python function + ek schema jo model ko batata hai ki kab aur kaise use kare.
def get_weather(city: str) -> dict:
    return {"city": city, "temp_c": 31, "sky": "haze"}  # asli app mein yahan weather API call hogi


def split_bill(total: float, people: int) -> dict:
    if people < 1:
        raise ValueError("people kam se kam 1 hona chahiye")
    return {"each": round(total / people, 2)}


TOOLS = {
    "get_weather": (get_weather, {
        "name": "get_weather",
        "description": "Kisi shehar ka abhi ka mausam (temperature, aasmaan). Aaj/abhi ke mausam ke liye; forecast ke liye nahi.",
        "input_schema": {"type": "object", "properties": {"city": {"type": "string"}},
                         "required": ["city"], "additionalProperties": False},
    }),
    "split_bill": (split_bill, {
        "name": "split_bill",
        "description": "Bill ko logon mein barabar baanto (rupaye mein).",
        "input_schema": {"type": "object",
                         "properties": {"total": {"type": "number"}, "people": {"type": "integer"}},
                         "required": ["total", "people"], "additionalProperties": False},
    }),
}


# 2) Tool chalana: galti ho toh crash nahi, model ko padhne layak error wapas.
def run_tool(call: dict) -> dict:
    result = {"type": "tool_result", "tool_use_id": call["id"]}
    if call["name"] not in TOOLS:
        return {**result, "content": f"unknown tool {call['name']}", "is_error": True}
    fn, spec = TOOLS[call["name"]]
    schema, args = spec["input_schema"], call["input"]
    missing = [k for k in schema["required"] if k not in args]
    unknown = [k for k in args if k not in schema["properties"]]
    if missing or unknown:  # chalane se pehle check: model ke arguments par bharosa nahi
        return {**result, "content": f"bad arguments: missing={missing} unknown={unknown}", "is_error": True}
    try:
        return {**result, "content": json.dumps(fn(**args))}
    except Exception as e:  # galat argument ya tool ki galti: model agle step mein sudhaar sakta hai
        return {**result, "content": f"{type(e).__name__}: {e}", "is_error": True}


# 3) Agent loop: model se poochho, jo tools maange woh chalao, results wapas bhejo, dohrao.
def run_agent(call_model, question: str, max_steps: int = 6) -> str:
    messages = [{"role": "user", "content": question}]
    specs = [spec for _, spec in TOOLS.values()]
    for _ in range(max_steps):  # step cap: agent kabhi bina ruke na chale
        reply = call_model(messages, specs)
        if reply["stop_reason"] == "max_tokens":
            return "Ruk gaya: jawab beech mein kat gaya (max_tokens)."
        if reply["stop_reason"] == "refusal":
            return "Ruk gaya: model ne mana kar diya."
        messages.append({"role": "assistant", "content": reply["content"]})
        calls = [b for b in reply["content"] if b["type"] == "tool_use"]
        if not calls:  # koi tool nahi maanga: yahi final jawab hai
            return "".join(b["text"] for b in reply["content"] if b["type"] == "text")
        messages.append({"role": "user", "content": [run_tool(c) for c in calls]})  # saare results ek saath
    return "Ruk gaya: bahut zyada tool steps."


# Asli model (Anthropic SDK; ANTHROPIC_API_KEY environment variable mein):
# from anthropic import Anthropic
# client = Anthropic()
# def claude(messages, specs):
#     r = client.messages.create(model=MODEL, max_tokens=1024, tools=specs, messages=messages)
#     return r.model_dump(exclude_none=True)

# Offline demo ke liye ek nakli model, taaki code bina API key ke chal jaaye:
def fake_model(messages, specs):
    last = messages[-1]["content"]
    if isinstance(last, str):
        return {"stop_reason": "tool_use", "content": [
            {"type": "tool_use", "id": "a1", "name": "get_weather", "input": {"city": "Mumbai"}},
            {"type": "tool_use", "id": "a2", "name": "split_bill", "input": {"total": 2400, "people": 4}},
        ]}
    r = {x["tool_use_id"]: json.loads(x["content"]) for x in last}
    return {"stop_reason": "end_turn", "content": [{"type": "text", "text":
            f"Mumbai: {r['a1']['temp_c']}°C, {r['a1']['sky']}. Har ek ke hisse {r['a2']['each']:g} rupaye."}]}


print(run_agent(fake_model, "Mumbai ka mausam kaisa hai? Aur 2400 ka bill 4 logon mein kitna aayega?"))
print(run_tool({"id": "t9", "name": "split_bill", "input": {"total": 100, "people": 0}}))
print(run_tool({"id": "t10", "name": "split_bill", "input": {"total": 100}}))

# Mumbai: 31°C, haze. Har ek ke hisse 600 rupaye.
# {'type': 'tool_result', 'tool_use_id': 't9', 'content': 'ValueError: people kam se kam 1 hona chahiye', 'is_error': True}
# {'type': 'tool_result', 'tool_use_id': 't10', 'content': "bad arguments: missing=['people'] unknown=[]", 'is_error': True}

agent.py (Python 3.9+; sirf standard library, asli model ke liye anthropic package). Chalaiye: python agent.py. Aakhri comment lines expected output hain.