gateway/modules/services/serviceAi/subWebResearch.py
2025-10-23 14:32:24 +02:00

389 lines
20 KiB
Python

import logging
from typing import Dict, Any, List, Optional, Tuple, Union
from modules.aicore.aicorePluginTavily import WebResearchRequest, WebResearchResult
from modules.interfaces.interfaceAiObjects import AiObjects
from modules.shared.configuration import APP_CONFIG
logger = logging.getLogger(__name__)
class SubWebResearch:
"""Web research operations including search, crawling, and analysis."""
def __init__(self, services, aiObjects):
"""Initialize web research service.
Args:
services: Service center instance for accessing other services
aiObjects: Initialized AiObjects instance
"""
self.services = services
self.aiObjects = aiObjects
async def webResearch(self, request: WebResearchRequest) -> WebResearchResult:
"""Perform web research using interface functions."""
try:
logger.info(f"WEB RESEARCH STARTED")
logger.info(f"User Query: {request.user_prompt}")
logger.info(f"Max Results: {request.max_results}, Max Pages: {request.options.max_pages}")
# Global URL index to track all processed URLs across the entire research session
global_processed_urls = set()
# Step 1: Find relevant websites - either provided URLs or AI-determined main URLs
logger.info(f"=== STEP 1: INITIAL MAIN URLS LIST ===")
if request.urls:
# Use provided URLs as initial main URLs
websites = request.urls
logger.info(f"Using provided URLs ({len(websites)}):")
for i, url in enumerate(websites, 1):
logger.info(f" {i}. {url}")
else:
# Use AI to determine main URLs based on user's intention
logger.info(f"AI analyzing user intent: '{request.user_prompt}'")
# Use AI to generate optimized Tavily search query and search parameters
query_optimizer_prompt = f"""You are a search query optimizer.
USER QUERY: {request.user_prompt}
Your task: Create a search query and parameters for the USER QUERY given.
RULES:
1. The search query MUST be related to the user query above
2. Extract key terms from the user query
3. Determine appropriate country/language based on the query context
4. Keep search query short (2-6 words)
Return ONLY this JSON format:
{{
"user_prompt": "search query based on user query above",
"country": "Full English country name (ISO-3166; map codes via pycountry/i18n-iso-countries)",
"language": "language_code_or_null",
"topic": "general|news|academic_or_null",
"time_range": "d|w|m|y_or_null",
"selection_strategy": "single|multiple|specific_page",
"selection_criteria": "what URLs to prioritize",
"expected_url_patterns": ["pattern1", "pattern2"],
"estimated_result_count": number
}}"""
# Get AI response for query optimization
from modules.datamodels.datamodelAi import AiCallRequest, AiCallOptions
ai_request = AiCallRequest(
prompt=query_optimizer_prompt,
options=AiCallOptions()
)
# Write web research query optimization prompt to debug file
self.services.utils.writeDebugFile(query_optimizer_prompt, "web_research_query_optimizer_prompt")
ai_response_obj = await self.aiObjects.call(ai_request)
ai_response = ai_response_obj.content
# Write web research query optimization response to debug file
self.services.utils.writeDebugFile(ai_response, "web_research_query_optimizer_response")
logger.debug(f"AI query optimizer response: {ai_response}")
# Parse AI response to extract search query
import json
try:
# Clean the response by removing markdown code blocks
cleaned_response = ai_response.strip()
if cleaned_response.startswith('```json'):
cleaned_response = cleaned_response[7:] # Remove ```json
if cleaned_response.endswith('```'):
cleaned_response = cleaned_response[:-3] # Remove ```
cleaned_response = cleaned_response.strip()
query_data = json.loads(cleaned_response)
search_query = query_data.get("user_prompt", request.user_prompt)
ai_country = query_data.get("country")
ai_language = query_data.get("language")
ai_topic = query_data.get("topic")
ai_time_range = query_data.get("time_range")
selection_strategy = query_data.get("selection_strategy", "multiple")
selection_criteria = query_data.get("selection_criteria", "relevant URLs")
expected_patterns = query_data.get("expected_url_patterns", [])
estimated_count = query_data.get("estimated_result_count", request.max_results)
logger.info(f"AI optimized search query: '{search_query}'")
logger.info(f"Selection strategy: {selection_strategy}")
logger.info(f"Selection criteria: {selection_criteria}")
logger.info(f"Expected URL patterns: {expected_patterns}")
logger.info(f"Estimated result count: {estimated_count}")
except json.JSONDecodeError:
logger.warning("Failed to parse AI response as JSON, using original query")
search_query = request.user_prompt
ai_country = None
ai_language = None
ai_topic = None
ai_time_range = None
selection_strategy = "multiple"
# Perform the web search with AI-determined parameters
search_kwargs = {
"query": search_query,
"max_results": request.max_results,
"search_depth": request.options.search_depth,
"auto_parameters": False # Use explicit parameters
}
# Add parameters only if they have valid values
def _normalizeCountry(c: Optional[str]) -> Optional[str]:
if not c:
return None
s = str(c).strip()
if not s or s.lower() in ['null', 'none', 'undefined']:
return None
# Map common codes to full English names when easy to do without extra deps
mapping = {
'ch': 'Switzerland', 'che': 'Switzerland',
'de': 'Germany', 'ger': 'Germany', 'deu': 'Germany',
'at': 'Austria', 'aut': 'Austria',
'us': 'United States', 'usa': 'United States', 'uni ted states': 'United States',
'uk': 'United Kingdom', 'gb': 'United Kingdom', 'gbr': 'United Kingdom'
}
key = s.lower()
if key in mapping:
return mapping[key]
# If looks like full name, capitalize first letter only (Tavily accepts English names)
return s
norm_ai_country = _normalizeCountry(ai_country)
norm_req_country = _normalizeCountry(request.options.country)
if norm_ai_country:
search_kwargs["country"] = norm_ai_country
elif norm_req_country:
search_kwargs["country"] = norm_req_country
if ai_language and ai_language not in ['null', '', 'none', 'undefined']:
search_kwargs["language"] = ai_language
elif request.options.language and request.options.language not in ['null', '', 'none', 'undefined']:
search_kwargs["language"] = request.options.language
if ai_topic and ai_topic in ['general', 'news', 'academic']:
search_kwargs["topic"] = ai_topic
elif request.options.topic and request.options.topic in ['general', 'news', 'academic']:
search_kwargs["topic"] = request.options.topic
if ai_time_range and ai_time_range in ['d', 'w', 'm', 'y']:
search_kwargs["time_range"] = ai_time_range
elif request.options.time_range and request.options.time_range in ['d', 'w', 'm', 'y']:
search_kwargs["time_range"] = request.options.time_range
# Constrain by expected domains if provided by AI
try:
include_domains = []
for p in expected_patterns or []:
if not isinstance(p, str):
continue
# Extract bare domain from pattern or URL
import re
m = re.search(r"(?:https?://)?([^/\s]+)", p.strip())
if m:
domain = m.group(1).lower()
# strip leading www.
if domain.startswith('www.'):
domain = domain[4:]
include_domains.append(domain)
# Deduplicate
if include_domains:
seen = set()
uniq = []
for d in include_domains:
if d not in seen:
seen.add(d)
uniq.append(d)
search_kwargs["include_domains"] = uniq
except Exception:
pass
# Log the parameters being used
logger.info(f"Search parameters: country={search_kwargs.get('country', 'not_set')}, language={search_kwargs.get('language', 'not_set')}, topic={search_kwargs.get('topic', 'not_set')}, time_range={search_kwargs.get('time_range', 'not_set')}, include_domains={search_kwargs.get('include_domains', [])}")
search_results = await self.aiObjects.search_websites(**search_kwargs)
logger.debug(f"Web search returned {len(search_results)} results:")
for i, result in enumerate(search_results, 1):
logger.debug(f" {i}. {result.url} - {result.title}")
# Deduplicate while preserving order
seen = set()
search_urls = []
for r in search_results:
u = str(r.url)
if u not in seen:
seen.add(u)
search_urls.append(u)
logger.info(f"After initial deduplication: {len(search_urls)} unique URLs from {len(search_results)} search results")
if not search_urls:
logger.error("No relevant websites found")
return WebResearchResult(success=False, error="No relevant websites found")
# Now use AI to determine the main URLs based on user's intention
logger.info(f"AI selecting main URLs from {len(search_urls)} search results based on user intent")
# Create a prompt for AI to identify main URLs based on user's intention
ai_prompt = f"""
Select the most relevant URLs from these search results:
{chr(10).join([f"{i+1}. {url}" for i, url in enumerate(search_urls)])}
Return only the URLs that are most relevant for the user's query.
One URL per line.
"""
# Create AI call request
ai_request = AiCallRequest(
prompt=ai_prompt,
options=AiCallOptions()
)
# Write web research URL selection prompt to debug file
self.services.utils.writeDebugFile(ai_prompt, "web_research_url_selection_prompt")
ai_response_obj = await self.aiObjects.call(ai_request)
ai_response = ai_response_obj.content
# Write web research URL selection response to debug file
self.services.utils.writeDebugFile(ai_response, "web_research_url_selection_response")
logger.debug(f"AI response for main URL selection: {ai_response}")
# Parse AI response to extract URLs
websites = []
for line in ai_response.strip().split('\n'):
line = line.strip()
if line and ('http://' in line or 'https://' in line):
# Extract URL from the line
for word in line.split():
if word.startswith('http://') or word.startswith('https://'):
websites.append(word.rstrip('.,;'))
break
if not websites:
logger.warning("AI did not identify any main URLs, using first few search results")
websites = search_urls[:3] # Fallback to first 3 search results
# Deduplicate while preserving order
seen = set()
unique_websites = []
for url in websites:
if url not in seen:
seen.add(url)
unique_websites.append(url)
websites = unique_websites
logger.info(f"After AI selection deduplication: {len(websites)} unique URLs from {len(websites)} AI-selected URLs")
logger.info(f"AI selected {len(websites)} main URLs (after deduplication):")
for i, url in enumerate(websites, 1):
logger.info(f" {i}. {url}")
# Step 2: Smart website selection using AI interface
logger.info(f"=== STEP 2: FILTERED URL LIST BY USER PROMPT'S INTENTION ===")
logger.info(f"AI analyzing {len(websites)} URLs for relevance to: '{request.user_prompt}'")
selectedWebsites, aiResponse = await self.aiObjects.selectRelevantWebsites(websites, request.user_prompt)
logger.debug(f"AI Response: {aiResponse}")
logger.debug(f"AI selected {len(selectedWebsites)} most relevant URLs:")
for i, url in enumerate(selectedWebsites, 1):
logger.debug(f" {i}. {url}")
# Show which were filtered out
filtered_out = [url for url in websites if url not in selectedWebsites]
if filtered_out:
logger.debug(f"Filtered out {len(filtered_out)} less relevant URLs:")
for i, url in enumerate(filtered_out, 1):
logger.debug(f" {i}. {url}")
# Step 3+4+5: Recursive crawling with configurable depth
# Get configuration parameters
max_depth = int(APP_CONFIG.get("Web_Research_MAX_DEPTH", "2"))
max_links_per_domain = int(APP_CONFIG.get("Web_Research_MAX_LINKS_PER_DOMAIN", "4"))
crawl_timeout_minutes = int(APP_CONFIG.get("Web_Research_CRAWL_TIMEOUT_MINUTES", "10"))
crawl_timeout_seconds = crawl_timeout_minutes * 60
# Use the configured max_depth or the request's pages_search_depth, whichever is smaller
effective_depth = min(max_depth, request.options.pages_search_depth)
logger.info(f"=== STEP 3+4+5: RECURSIVE CRAWLING (DEPTH {effective_depth}) ===")
logger.info(f"Starting recursive crawl of {len(selectedWebsites)} main websites...")
logger.info(f"Search depth: {effective_depth} levels (max configured: {max_depth})")
logger.info(f"Max links per domain: {max_links_per_domain}")
logger.info(f"Crawl timeout: {crawl_timeout_minutes} minutes")
# Use recursive crawling with URL index to avoid duplicates
import asyncio
try:
allContent = await asyncio.wait_for(
self.aiObjects.crawlRecursively(
urls=selectedWebsites,
max_depth=effective_depth,
extract_depth=request.options.extract_depth,
max_per_domain=max_links_per_domain,
global_processed_urls=global_processed_urls
),
timeout=crawl_timeout_seconds
)
logger.info(f"Crawling completed within timeout: {len(allContent)} pages crawled")
except asyncio.TimeoutError:
logger.warning(f"Crawling timed out after {crawl_timeout_minutes} minutes, using partial results")
# crawlRecursively now handles timeouts gracefully and returns partial results
# Try to get the partial results that were collected
allContent = {}
if not allContent:
logger.error("Could not extract content from any websites")
return WebResearchResult(success=False, error="Could not extract content from any websites")
logger.info(f"=== WEB RESEARCH COMPLETED ===")
logger.info(f"Successfully crawled {len(allContent)} URLs total")
logger.info(f"Crawl depth: {effective_depth} levels")
# Create simple result with raw content
sources = [{"title": url, "url": url} for url in selectedWebsites]
# Get all additional links (all URLs except main ones)
additional_links = [url for url in allContent.keys() if url not in selectedWebsites]
# Combine all content into a single result
combinedContent = ""
for url, content in allContent.items():
combinedContent += f"\n\n=== {url} ===\n{content}\n"
# Create simplified document structure
document = {
"documentName": f"webResearch_{request.user_prompt[:50]}.json",
"documentData": {
"user_prompt": request.user_prompt,
"analysis_result": combinedContent,
"sources": sources,
"additional_links": additional_links,
"metadata": {
"websites_analyzed": len(allContent),
"additional_links_found": len(additional_links),
"crawl_depth": effective_depth,
"max_configured_depth": max_depth,
"max_links_per_domain": max_links_per_domain,
"crawl_timeout_minutes": crawl_timeout_minutes,
"total_urls_crawled": len(allContent),
"main_urls": len(selectedWebsites),
"additional_urls": len(additional_links)
}
},
"mimeType": "application/json"
}
return WebResearchResult(
success=True,
documents=[document]
)
except Exception as e:
logger.error(f"Error in web research: {str(e)}")
return WebResearchResult(success=False, error=str(e))