app.py

from __future__ import annotations

import ast
import ipaddress
import operator
import os
import re
import socket
import threading
from datetime import datetime
from pathlib import Path
from typing import Any
from urllib.parse import urljoin, urlparse

from dotenv import load_dotenv


load_dotenv()

PERSISTENT_ROOT = Path(os.getenv("PERSISTENT_ROOT", "/data")).expanduser()
try:
    PERSISTENT_ROOT.mkdir(parents=True, exist_ok=True)
except OSError:
    PERSISTENT_ROOT = Path("data")
    PERSISTENT_ROOT.mkdir(parents=True, exist_ok=True)

os.environ.setdefault("HF_HOME", str(PERSISTENT_ROOT / "huggingface"))
os.environ.setdefault("HF_HUB_CACHE", str(PERSISTENT_ROOT / "huggingface" / "hub"))

import gradio as gr
import requests
from bs4 import BeautifulSoup
from ddgs import DDGS
from huggingface_hub import hf_hub_download
from llama_cpp import Llama
from smolagents import ChatMessage, CodeAgent, Model, Tool


MODEL_REPO = os.getenv("MODEL_REPO", "NANI-Nithin/K2-Horizon-0.9B-GGUF")
MODEL_FILE = os.getenv("MODEL_FILE", "K2-Horizon-0.9B-Q4_K_M.gguf")
MODEL_DIR = Path(os.getenv("MODEL_DIR", str(PERSISTENT_ROOT / "models"))).expanduser()


def env_int(name: str, default: int) -> int:
    try:
        return int(os.getenv(name, str(default)))
    except ValueError:
        return default


def env_float(name: str, default: float) -> float:
    try:
        return float(os.getenv(name, str(default)))
    except ValueError:
        return default


def content_to_text(content: Any) -> str:
    if isinstance(content, str):
        return content
    if isinstance(content, list):
        parts: list[str] = []
        for item in content:
            if isinstance(item, dict):
                parts.append(str(item.get("text", item.get("content", ""))))
            else:
                parts.append(str(item))
        return "\n".join(part for part in parts if part)
    return str(content or "")


class LlamaCppModel(Model):
    def __init__(self, llama: Llama, max_tokens: int, temperature: float) -> None:
        super().__init__()
        self.llama = llama
        self.max_tokens = max_tokens
        self.temperature = temperature

    @staticmethod
    def _normalize_messages(messages: list[Any]) -> list[dict[str, str]]:
        normalized: list[dict[str, str]] = []
        for message in messages:
            if isinstance(message, dict):
                role = message.get("role", "user")
                content = message.get("content", "")
            else:
                role = getattr(message, "role", "user")
                content = getattr(message, "content", "")
            if hasattr(role, "value"):
                role = role.value
            role = str(role).lower()
            if role not in {"system", "user", "assistant"}:
                role = "user"
            normalized.append({"role": role, "content": content_to_text(content)})
        return normalized

    def generate(
        self,
        messages: list[Any],
        stop_sequences: list[str] | None = None,
        response_format: dict[str, Any] | None = None,
        tools_to_call_from: list[Tool] | None = None,
        **kwargs: Any,
    ) -> ChatMessage:
        del response_format, tools_to_call_from
        result = self.llama.create_chat_completion(
            messages=self._normalize_messages(messages),
            max_tokens=int(kwargs.get("max_tokens", self.max_tokens)),
            temperature=float(kwargs.get("temperature", self.temperature)),
            top_p=float(kwargs.get("top_p", 0.9)),
            repeat_penalty=float(kwargs.get("repeat_penalty", 1.1)),
            stop=stop_sequences or None,
        )
        content = result["choices"][0]["message"].get("content", "")
        return ChatMessage(role="assistant", content=content)

    def __call__(self, messages: list[Any], **kwargs: Any) -> ChatMessage:
        return self.generate(messages, **kwargs)

    def direct_chat(self, messages: list[dict[str, str]]) -> str:
        result = self.llama.create_chat_completion(
            messages=messages,
            max_tokens=self.max_tokens,
            temperature=self.temperature,
            top_p=0.9,
            repeat_penalty=1.1,
        )
        return str(result["choices"][0]["message"].get("content", "")).strip()


class DuckDuckGoSearchTool(Tool):
    name = "web_search"
    description = "Search the public web with DuckDuckGo. Use it for current facts and external information."
    inputs = {
        "query": {"type": "string", "description": "A focused web search query."},
        "max_results": {
            "type": "integer",
            "description": "Number of results from 1 to 8.",
            "nullable": True,
        },
    }
    output_type = "string"

    def forward(self, query: str, max_results: int | None = None) -> str:
        limit = max(1, min(int(max_results or 5), 8))
        results = list(DDGS().text(query, max_results=limit))
        if not results:
            return "No search results found."
        rows = []
        for index, item in enumerate(results, 1):
            title = item.get("title", "Untitled")
            url = item.get("href", item.get("url", ""))
            body = item.get("body", "")
            rows.append(f"{index}. {title}\nURL: {url}\nSnippet: {body}")
        return "\n\n".join(rows)


def ensure_public_url(url: str) -> str:
    parsed = urlparse(url)
    if parsed.scheme not in {"http", "https"} or not parsed.hostname:
        raise ValueError("Only public http/https URLs are allowed.")
    addresses = socket.getaddrinfo(parsed.hostname, parsed.port or 80, proto=socket.IPPROTO_TCP)
    for address in addresses:
        ip = ipaddress.ip_address(address[4][0])
        if not ip.is_global:
            raise ValueError("Private, loopback, and local network addresses are blocked.")
    return url


class ReadWebpageTool(Tool):
    name = "read_webpage"
    description = "Download and extract readable text from a public web page URL."
    inputs = {
        "url": {"type": "string", "description": "The full public http or https URL."},
    }
    output_type = "string"

    def forward(self, url: str) -> str:
        current_url = url
        response = None
        for _ in range(4):
            safe_url = ensure_public_url(current_url)
            response = requests.get(
                safe_url,
                timeout=12,
                allow_redirects=False,
                headers={"User-Agent": "Mozilla/5.0 (compatible; K2-Horizon-Agent/1.0)"},
            )
            if response.status_code not in {301, 302, 303, 307, 308}:
                break
            location = response.headers.get("location")
            if not location:
                break
            current_url = urljoin(current_url, location)
        assert response is not None
        response.raise_for_status()
        content_type = response.headers.get("content-type", "")
        if "text/html" not in content_type and "text/plain" not in content_type:
            return f"Unsupported content type: {content_type}"
        soup = BeautifulSoup(response.text[:2_000_000], "html.parser")
        for node in soup(["script", "style", "noscript", "svg"]):
            node.decompose()
        text = re.sub(r"\n{3,}", "\n\n", soup.get_text("\n", strip=True))
        return text[:12_000] or "No readable text found."


_BINARY_OPERATORS = {
    ast.Add: operator.add,
    ast.Sub: operator.sub,
    ast.Mult: operator.mul,
    ast.Div: operator.truediv,
    ast.FloorDiv: operator.floordiv,
    ast.Mod: operator.mod,
    ast.Pow: operator.pow,
}
_UNARY_OPERATORS = {ast.UAdd: operator.pos, ast.USub: operator.neg}


def evaluate_expression(node: ast.AST) -> float | int:
    if isinstance(node, ast.Expression):
        return evaluate_expression(node.body)
    if isinstance(node, ast.Constant) and isinstance(node.value, (int, float)):
        return node.value
    if isinstance(node, ast.BinOp) and type(node.op) in _BINARY_OPERATORS:
        left = evaluate_expression(node.left)
        right = evaluate_expression(node.right)
        if isinstance(node.op, ast.Pow) and abs(right) > 100:
            raise ValueError("Exponent is too large.")
        return _BINARY_OPERATORS[type(node.op)](left, right)
    if isinstance(node, ast.UnaryOp) and type(node.op) in _UNARY_OPERATORS:
        return _UNARY_OPERATORS[type(node.op)](evaluate_expression(node.operand))
    raise ValueError("Only numeric arithmetic is supported.")


class CalculatorTool(Tool):
    name = "calculator"
    description = "Safely evaluate a numeric arithmetic expression."
    inputs = {"expression": {"type": "string", "description": "Arithmetic expression to evaluate."}}
    output_type = "string"

    def forward(self, expression: str) -> str:
        if len(expression) > 200:
            raise ValueError("Expression is too long.")
        value = evaluate_expression(ast.parse(expression, mode="eval"))
        return str(value)


class CurrentTimeTool(Tool):
    name = "current_time"
    description = "Get the current system date and time, including timezone."
    inputs = {}
    output_type = "string"

    def forward(self) -> str:
        return datetime.now().astimezone().isoformat(timespec="seconds")


class AppRuntime:
    def __init__(self) -> None:
        self.lock = threading.Lock()
        self.model: LlamaCppModel | None = None
        self.agent: CodeAgent | None = None
        self.model_path: Path | None = None

    def load(self) -> None:
        if self.model is not None:
            return
        with self.lock:
            if self.model is not None:
                return
            MODEL_DIR.mkdir(parents=True, exist_ok=True)
            local_path = hf_hub_download(
                repo_id=MODEL_REPO,
                filename=MODEL_FILE,
                local_dir=str(MODEL_DIR),
            )
            self.model_path = Path(local_path)
            llama = Llama(
                model_path=str(self.model_path),
                n_ctx=env_int("N_CTX", 4096),
                n_threads=env_int("N_THREADS", max(1, (os.cpu_count() or 4) - 1)),
                n_threads_batch=env_int("N_THREADS_BATCH", os.cpu_count() or 4),
                n_batch=env_int("N_BATCH", 256),
                n_gpu_layers=0,
                use_mmap=True,
                verbose=os.getenv("LLAMA_VERBOSE", "0") == "1",
            )
            self.model = LlamaCppModel(
                llama=llama,
                max_tokens=env_int("MAX_NEW_TOKENS", 700),
                temperature=env_float("TEMPERATURE", 0.2),
            )
            self.agent = CodeAgent(
                tools=[
                    DuckDuckGoSearchTool(),
                    ReadWebpageTool(),
                    CalculatorTool(),
                    CurrentTimeTool(),
                ],
                model=self.model,
                max_steps=env_int("AGENT_MAX_STEPS", 5),
                add_base_tools=False,
                additional_authorized_imports=[],
                code_block_tags="markdown",
            )

    def reply(self, message: str, history: list[dict[str, str]], use_tools: bool) -> str:
        self.load()
        assert self.model is not None
        if use_tools:
            assert self.agent is not None
            transcript = "\n".join(
                f"{item.get('role', 'user')}: {content_to_text(item.get('content', ''))}"
                for item in history[-6:]
                if item.get("role") in {"user", "assistant"}
            )
            task = message
            if transcript:
                task = f"Conversation context:\n{transcript}\n\nCurrent user request:\n{message}"
            return str(self.agent.run(task, reset=True)).strip()
        system = {
            "role": "system",
            "content": "You are K2 Horizon, a concise and helpful local assistant.",
        }
        context = [system]
        for item in history[-10:]:
            if item.get("role") in {"user", "assistant"}:
                context.append({"role": item["role"], "content": content_to_text(item.get("content", ""))})
        context.append({"role": "user", "content": message})
        return self.model.direct_chat(context)


runtime = AppRuntime()


def respond(message: str, history: list[dict[str, str]], use_tools: bool):
    if not message.strip():
        return "", history
    updated = history + [{"role": "user", "content": message}]
    try:
        answer = runtime.reply(message.strip(), history, use_tools)
    except Exception as exc:
        answer = f"Error: {type(exc).__name__}: {exc}"
    updated.append({"role": "assistant", "content": answer})
    return "", updated


CSS = """
.gradio-container { max-width: 860px !important; margin: 0 auto !important; }
#app-shell { min-height: 100vh; padding: 32px 12px 20px; }
#title { text-align: center; margin-bottom: 2px; }
#subtitle { text-align: center; color: var(--body-text-color-subdued); margin-bottom: 18px; }
#chat { border: 1px solid var(--border-color-primary); border-radius: 18px; overflow: hidden; }
#composer { gap: 10px; align-items: stretch; margin-top: 12px; }
#prompt textarea { border-radius: 14px !important; }
#send { min-width: 92px; border-radius: 14px !important; }
#controls { align-items: center; margin-top: 8px; }
#note { color: var(--body-text-color-subdued); font-size: 12px; text-align: right; }
footer { display: none !important; }
"""

THEME = gr.themes.Base(primary_hue="slate", neutral_hue="slate")


with gr.Blocks(
    title="K2 Horizon",
) as demo:
    with gr.Column(elem_id="app-shell"):
        gr.Markdown("# K2 Horizon", elem_id="title")
        gr.Markdown("Private CPU inference with optional web tools", elem_id="subtitle")
        chatbot = gr.Chatbot(
            height=570,
            buttons=["copy"],
            allow_tags=False,
            placeholder="Ask anything",
            elem_id="chat",
        )
        with gr.Row(elem_id="composer"):
            prompt = gr.Textbox(
                placeholder="Message K2 Horizon…",
                show_label=False,
                scale=9,
                lines=1,
                max_lines=5,
                elem_id="prompt",
            )
            send = gr.Button("Send", variant="primary", scale=1, elem_id="send")
        with gr.Row(elem_id="controls"):
            use_tools = gr.Checkbox(value=True, label="Web tools", scale=1)
            clear = gr.Button("Clear", variant="secondary", size="sm", scale=0)
            gr.Markdown("Model loads on the first message", elem_id="note")

    send.click(respond, [prompt, chatbot, use_tools], [prompt, chatbot])
    prompt.submit(respond, [prompt, chatbot, use_tools], [prompt, chatbot])
    clear.click(lambda: ("", []), outputs=[prompt, chatbot], queue=False)


if __name__ == "__main__":
    demo.queue(default_concurrency_limit=1).launch(
        server_name=os.getenv("GRADIO_SERVER_NAME", "0.0.0.0"),
        server_port=env_int("GRADIO_SERVER_PORT", 7860),
        share=os.getenv("GRADIO_SHARE", "0") == "1",
        show_error=True,
        ssr_mode=False,
        footer_links=[],
        theme=THEME,
        css=CSS,
    )