246 lines
7.6 KiB
Python
246 lines
7.6 KiB
Python
"""Búsqueda web para Arisu — DuckDuckGo, Wikipedia y Serper (opcional)."""
|
|
|
|
import os
|
|
import re
|
|
from datetime import datetime
|
|
from zoneinfo import ZoneInfo
|
|
|
|
import requests as http_requests
|
|
from duckduckgo_search import DDGS
|
|
|
|
UA = "ArisuAI/1.0 (Educational assistant; +local)"
|
|
MAX_RESULTS = 5
|
|
SNIPPET_LEN = 320
|
|
TIMEOUT = 12
|
|
|
|
_WIKI_LANG = {"es": "es", "en": "en", "ja": "ja", "fr": "fr", "de": "de", "pt": "pt", "ko": "ko", "zh": "zh"}
|
|
_DAYS_ES = ("lunes", "martes", "miércoles", "jueves", "viernes", "sábado", "domingo")
|
|
_MONTHS_ES = (
|
|
"enero", "febrero", "marzo", "abril", "mayo", "junio",
|
|
"julio", "agosto", "septiembre", "octubre", "noviembre", "diciembre",
|
|
)
|
|
_RELATIVE_WORDS = (
|
|
"hoy", "today", "ahora", "now", "actual", "actualidad", "reciente",
|
|
"últim", "ultim", "esta semana", "this week", "este mes", "this month",
|
|
"en vivo", "breaking", "noticias de",
|
|
)
|
|
|
|
|
|
def _now() -> datetime:
|
|
tz = os.getenv("APP_TIMEZONE", "America/Mexico_City")
|
|
try:
|
|
return datetime.now(ZoneInfo(tz))
|
|
except Exception:
|
|
return datetime.now()
|
|
|
|
|
|
def current_date_context(lang: str = "es") -> str:
|
|
now = _now()
|
|
iso = now.strftime("%Y-%m-%d")
|
|
year = now.year
|
|
if lang.startswith("es"):
|
|
human = f"{_DAYS_ES[now.weekday()]}, {now.day} de {_MONTHS_ES[now.month - 1]} de {year}"
|
|
else:
|
|
human = now.strftime("%A, %B %d, %Y")
|
|
return (
|
|
f"FECHA Y HORA ACTUAL DEL SISTEMA (autoritativa — NO uses tu fecha de entrenamiento):\n"
|
|
f"- Hoy es: {human}\n"
|
|
f"- ISO: {iso}\n"
|
|
f"- Año en curso: {year}\n"
|
|
f"- «Hoy» / «today» / «actual» = {iso}, año {year}."
|
|
)
|
|
|
|
|
|
def enrich_search_query(query: str) -> str:
|
|
"""Añade fecha explícita si la consulta es relativa al presente."""
|
|
q = (query or "").strip()
|
|
if not q:
|
|
return q
|
|
now = _now()
|
|
iso = now.strftime("%Y-%m-%d")
|
|
year = str(now.year)
|
|
lower = q.lower()
|
|
|
|
if year in q:
|
|
return q
|
|
|
|
if any(w in lower for w in _RELATIVE_WORDS):
|
|
return f"{q} {iso}"
|
|
|
|
return q
|
|
|
|
|
|
def _clip(text: str, n: int = SNIPPET_LEN) -> str:
|
|
text = re.sub(r"<[^>]+>", "", text or "").strip()
|
|
if len(text) <= n:
|
|
return text
|
|
return text[:n] + "…"
|
|
|
|
|
|
def _serper_search(query: str, api_key: str) -> str:
|
|
resp = http_requests.post(
|
|
"https://google.serper.dev/search",
|
|
json={"q": query, "num": MAX_RESULTS},
|
|
headers={"X-API-KEY": api_key, "Content-Type": "application/json"},
|
|
timeout=TIMEOUT,
|
|
)
|
|
resp.raise_for_status()
|
|
data = resp.json()
|
|
organic = data.get("organic") or []
|
|
if not organic:
|
|
return f"No se encontraron resultados para: {query}"
|
|
|
|
blocks = []
|
|
for i, r in enumerate(organic[:MAX_RESULTS], 1):
|
|
title = r.get("title", "")
|
|
link = r.get("link", "")
|
|
snippet = _clip(r.get("snippet", ""))
|
|
blocks.append(f"{i}. {title}\nURL: {link}\n{snippet}")
|
|
return "\n\n".join(blocks)
|
|
|
|
|
|
def _ddg_search(query: str) -> str:
|
|
with DDGS() as ddgs:
|
|
hits = list(ddgs.text(query, max_results=MAX_RESULTS))
|
|
if not hits:
|
|
return ""
|
|
blocks = []
|
|
for i, r in enumerate(hits, 1):
|
|
title = (r.get("title") or "").strip()
|
|
href = (r.get("href") or "").strip()
|
|
body = _clip(r.get("body") or "")
|
|
blocks.append(f"{i}. {title}\nURL: {href}\n{body}")
|
|
return "\n\n".join(blocks)
|
|
|
|
|
|
def _wiki_search(query: str, lang: str = "es") -> str:
|
|
wiki_lang = _WIKI_LANG.get(lang[:2].lower(), "en")
|
|
headers = {"User-Agent": UA}
|
|
base = f"https://{wiki_lang}.wikipedia.org/w/api.php"
|
|
|
|
sr = http_requests.get(
|
|
base,
|
|
params={
|
|
"action": "query",
|
|
"list": "search",
|
|
"srsearch": query,
|
|
"format": "json",
|
|
"srlimit": 3,
|
|
},
|
|
headers=headers,
|
|
timeout=TIMEOUT,
|
|
)
|
|
sr.raise_for_status()
|
|
hits = sr.json().get("query", {}).get("search") or []
|
|
if not hits:
|
|
return ""
|
|
|
|
blocks = []
|
|
for i, h in enumerate(hits, 1):
|
|
title = h.get("title", "")
|
|
snippet = _clip(h.get("snippet", ""))
|
|
page_url = f"https://{wiki_lang}.wikipedia.org/wiki/{title.replace(' ', '_')}"
|
|
|
|
extract = snippet
|
|
try:
|
|
er = http_requests.get(
|
|
base,
|
|
params={
|
|
"action": "query",
|
|
"prop": "extracts",
|
|
"exintro": True,
|
|
"explaintext": True,
|
|
"titles": title,
|
|
"format": "json",
|
|
},
|
|
headers=headers,
|
|
timeout=TIMEOUT,
|
|
)
|
|
er.raise_for_status()
|
|
pages = er.json().get("query", {}).get("pages") or {}
|
|
for p in pages.values():
|
|
if p.get("extract"):
|
|
extract = _clip(p["extract"], 400)
|
|
break
|
|
except Exception:
|
|
pass
|
|
|
|
blocks.append(f"{i}. {title} (Wikipedia)\nURL: {page_url}\n{extract}")
|
|
|
|
header = f"[Wikipedia — {wiki_lang}]\n"
|
|
return header + "\n\n".join(blocks)
|
|
|
|
|
|
def web_search(query: str, lang: str = "es") -> str:
|
|
raw_q = (query or "").strip()
|
|
if not raw_q:
|
|
return "Error: consulta de búsqueda vacía"
|
|
|
|
q = enrich_search_query(raw_q)
|
|
now = _now()
|
|
date_header = f"[Búsqueda — fecha del sistema: {now.strftime('%Y-%m-%d')}]\n"
|
|
|
|
errors: list[str] = []
|
|
|
|
serper_key = os.getenv("SERPER_API_KEY", "").strip()
|
|
if serper_key:
|
|
try:
|
|
result = _serper_search(q, serper_key)
|
|
if result and "No se encontraron" not in result:
|
|
note = f"Consulta enviada: {q}\n\n" if q != raw_q else ""
|
|
return date_header + note + f"[Google/Serper]\n{result}"
|
|
except Exception as e:
|
|
errors.append(f"Serper: {e}")
|
|
|
|
try:
|
|
ddg = _ddg_search(q)
|
|
if ddg:
|
|
note = f"Consulta enviada: {q}\n\n" if q != raw_q else ""
|
|
return date_header + note + f"[DuckDuckGo]\n{ddg}"
|
|
except Exception as e:
|
|
errors.append(f"DuckDuckGo: {e}")
|
|
|
|
try:
|
|
wiki = _wiki_search(q, lang)
|
|
if wiki:
|
|
note = f"Consulta enviada: {q}\n\n" if q != raw_q else ""
|
|
return date_header + note + wiki
|
|
except Exception as e:
|
|
errors.append(f"Wikipedia: {e}")
|
|
|
|
detail = "; ".join(errors) if errors else "sin resultados"
|
|
return f"No se pudo obtener información web para «{raw_q}». ({detail})"
|
|
|
|
|
|
WEB_SEARCH_TOOL = {
|
|
"type": "function",
|
|
"function": {
|
|
"name": "web_search",
|
|
"description": (
|
|
"Busca en internet cuando no sepas algo, necesites datos actuales, "
|
|
"noticias o hechos recientes. Usa la FECHA ACTUAL del system prompt: "
|
|
"si el usuario dice «hoy» o «actual», incluye el año en curso en la query."
|
|
),
|
|
"parameters": {
|
|
"type": "object",
|
|
"properties": {
|
|
"query": {
|
|
"type": "string",
|
|
"description": (
|
|
"Consulta clara. Si es sobre «hoy» o lo reciente, "
|
|
"incluye la fecha ISO o el año actual del system prompt."
|
|
),
|
|
},
|
|
},
|
|
"required": ["query"],
|
|
},
|
|
},
|
|
}
|
|
|
|
WEB_PROMPT = """
|
|
BÚSQUEDA WEB:
|
|
Si no estás segura o piden algo actual/reciente, usa web_search ANTES de inventar.
|
|
IMPORTANTE: la FECHA ACTUAL del system prompt es la verdad — NO uses 2024 ni tu corte de entrenamiento.
|
|
Si dicen «hoy», «actual», «reciente», la búsqueda debe usar el año y fecha del system prompt.
|
|
Tras buscar, resume en 2-3 oraciones con personalidad kawaii.
|
|
"""
|