import os import io import uuid import re from typing import Dict, Any from pathlib import Path import mimetypes import chardet import PyPDF2 import docx from core.utils.logger import logger from core.services.supabase import DBConnection from core.services.llm import make_llm_api_call class FileProcessor: SUPPORTED_EXTENSIONS = {'.txt', '.pdf', '.docx'} MAX_FILE_SIZE = 50 * 1024 * 1024 def __init__(self): self.db = DBConnection() def sanitize_filename(self, filename: str) -> str: """Sanitize filename for S3 storage - remove/replace invalid characters.""" # Keep the file extension name, ext = os.path.splitext(filename) # Replace emojis and other problematic Unicode characters with underscores name = re.sub(r'[^\w\s\-\.]', '_', name) # Replace spaces with underscores name = re.sub(r'\s+', '_', name) # Remove multiple consecutive underscores name = re.sub(r'_+', '_', name) # Remove leading/trailing underscores name = name.strip('_') # Ensure name is not empty if not name: name = 'file' return f"{name}{ext}" def _is_likely_text_file(self, file_content: bytes) -> bool: """Check if file content is likely text-based.""" try: # Try to decode as text detected = chardet.detect(file_content[:1024]) # Check first 1KB if detected.get('confidence', 0) > 0.7: decoded = file_content[:1024].decode(detected.get('encoding', 'utf-8')) # Check if most characters are printable printable_ratio = len([c for c in decoded if c.isprintable() or c.isspace()]) / len(decoded) return printable_ratio > 0.8 except: pass return False async def process_file_fast( self, account_id: str, folder_id: str, file_content: bytes, filename: str, mime_type: str, background_tasks ) -> Dict[str, Any]: """Fast file processing - upload file immediately, generate summary in background.""" import time start = time.time() logger.info(f"[PROCESSOR] Starting fast processing for: {filename}") try: if len(file_content) > self.MAX_FILE_SIZE: raise ValueError(f"File too large: {len(file_content)} bytes") file_extension = Path(filename).suffix.lower() # Check if it's text-based first is_text_based = ( mime_type.startswith('text/') or mime_type in ['application/json', 'application/xml', 'text/xml'] or self._is_likely_text_file(file_content) ) # If not text-based, check allowed extensions if not is_text_based and file_extension not in self.SUPPORTED_EXTENSIONS: raise ValueError(f"Unsupported file type: {file_extension}") # Generate unique entry ID entry_id = str(uuid.uuid4()) # Sanitize filename for S3 storage sanitized_filename = self.sanitize_filename(filename) # Upload to S3 s3_path = f"knowledge-base/{folder_id}/{entry_id}/{sanitized_filename}" client = await self.db.client t1 = time.time() await client.storage.from_('file-uploads').upload( s3_path, file_content, {"content-type": mime_type} ) logger.info(f"[PROCESSOR] S3 upload took: {time.time() - t1:.2f}s") # Save to database with placeholder summary entry_data = { 'entry_id': entry_id, 'folder_id': folder_id, 'account_id': account_id, 'filename': filename, 'file_path': s3_path, 'file_size': len(file_content), 'mime_type': mime_type, 'summary': 'Processing...', 'is_active': True } t2 = time.time() await client.table('knowledge_base_entries').insert(entry_data).execute() logger.info(f"[PROCESSOR] DB insert took: {time.time() - t2:.2f}s") # Schedule background summary generation t3 = time.time() background_tasks.add_task( self._generate_and_update_summary, entry_id, file_content, filename, mime_type ) logger.info(f"[PROCESSOR] Background task scheduled in: {time.time() - t3:.2f}s") logger.info(f"[PROCESSOR] Total fast processing time: {time.time() - start:.2f}s") return { 'success': True, 'entry_id': entry_id, 'filename': filename, 'processing': True } except Exception as e: logger.error(f"Error processing file {filename}: {str(e)}") return {'success': False, 'error': str(e)} async def _generate_and_update_summary( self, entry_id: str, file_content: bytes, filename: str, mime_type: str ): """Background task to generate and update file summary.""" try: # Extract content content = self._extract_content(file_content, filename, mime_type) if not content: content = f"File: {filename} ({len(file_content)} bytes, {mime_type})" # Generate summary summary = await self._generate_summary(content, filename) # Update database client = await self.db.client await client.table('knowledge_base_entries').update({ 'summary': summary }).eq('entry_id', entry_id).execute() logger.info(f"Successfully generated summary for entry {entry_id}") except Exception as e: logger.error(f"Error generating summary for entry {entry_id}: {str(e)}") # Update with error message try: client = await self.db.client await client.table('knowledge_base_entries').update({ 'summary': f"Error generating summary: {str(e)}" }).eq('entry_id', entry_id).execute() except: pass async def process_file( self, account_id: str, folder_id: str, file_content: bytes, filename: str, mime_type: str ) -> Dict[str, Any]: """Legacy synchronous file processing - kept for backward compatibility.""" try: if len(file_content) > self.MAX_FILE_SIZE: raise ValueError(f"File too large: {len(file_content)} bytes") file_extension = Path(filename).suffix.lower() # Check if it's text-based first is_text_based = ( mime_type.startswith('text/') or mime_type in ['application/json', 'application/xml', 'text/xml'] or self._is_likely_text_file(file_content) ) # If not text-based, check allowed extensions if not is_text_based and file_extension not in self.SUPPORTED_EXTENSIONS: raise ValueError(f"Unsupported file type: {file_extension}") # Generate unique entry ID entry_id = str(uuid.uuid4()) # Sanitize filename for S3 storage sanitized_filename = self.sanitize_filename(filename) # Upload to S3 s3_path = f"knowledge-base/{folder_id}/{entry_id}/{sanitized_filename}" client = await self.db.client await client.storage.from_('file-uploads').upload( s3_path, file_content, {"content-type": mime_type} ) # Extract content for summary content = self._extract_content(file_content, filename, mime_type) if not content: # If no content could be extracted, create a basic file info summary content = f"File: {filename} ({len(file_content)} bytes, {mime_type})" # Generate LLM summary summary = await self._generate_summary(content, filename) # Save to database entry_data = { 'entry_id': entry_id, 'folder_id': folder_id, 'account_id': account_id, 'filename': filename, 'file_path': s3_path, 'file_size': len(file_content), 'mime_type': mime_type, 'summary': summary, 'is_active': True } result = await client.table('knowledge_base_entries').insert(entry_data).execute() return { 'success': True, 'entry_id': entry_id, 'filename': filename, 'summary_length': len(summary) } except Exception as e: logger.error(f"Error processing file {filename}: {str(e)}") return {'success': False, 'error': str(e)} async def _generate_summary(self, content: str, filename: str) -> str: """Generate LLM summary of file content with smart chunking and fallbacks.""" try: # Model priority: Google Gemini → OpenRouter → GPT-5 Mini models = [ ("google/gemini-2.5-flash-lite", 1_000_000), # 1M context ("openrouter/google/gemini-2.5-flash-lite", 1_000_000), # Fallback ("gpt-5-mini", 400_000) # Final fallback ] # Estimate tokens (rough: 1 token ≈ 4 chars) estimated_tokens = len(content) // 4 for model_name, context_limit in models: try: # Reserve tokens for prompt and response usable_context = context_limit - 1000 # Reserve for prompt + response if estimated_tokens <= usable_context: # Content fits, use full content processed_content = content else: # Content too large, intelligent chunking processed_content = self._smart_chunk_content(content, usable_context * 4) # Convert back to chars prompt = f"""Analyze this file and create a concise, actionable summary for an AI agent's knowledge base. File: {filename} Content: {processed_content} Generate a 2-3 sentence summary that captures: 1. What this file contains 2. Key information or purpose 3. When this knowledge would be useful Keep it under 200 words and make it actionable for context injection.""" messages = [{"role": "user", "content": prompt}] response = await make_llm_api_call( messages=messages, model_name=model_name, temperature=0.1, max_tokens=300 ) summary = response.choices[0].message.content.strip() if summary: logger.info(f"Summary generated successfully using {model_name}") return summary except Exception as e: logger.warning(f"Model {model_name} failed: {str(e)}") continue # All models failed - high-reliability fallback logger.error("All LLM models failed, using intelligent fallback") return self._create_fallback_summary(content, filename) except Exception as e: logger.error(f"Error generating summary: {str(e)}") return self._create_fallback_summary(content, filename) def _smart_chunk_content(self, content: str, max_chars: int) -> str: """Intelligently chunk content to fit context limits.""" if len(content) <= max_chars: return content # Strategy: Take beginning + end with priority sections quarter = max_chars // 4 # Always include beginning (most important) beginning = content[:quarter * 2] # Include end for conclusion/summary ending = content[-quarter:] # Try to find important middle sections (headers, key terms) middle_section = "" remaining_chars = max_chars - len(beginning) - len(ending) - 100 # Buffer if remaining_chars > 0: middle_start = len(beginning) middle_end = len(content) - quarter middle_content = content[middle_start:middle_end] # Look for important sections (lines with caps, numbers, bullets) lines = middle_content.split('\n') important_lines = [] current_length = 0 for line in lines: line_stripped = line.strip() if not line_stripped: continue # Prioritize lines that look important is_important = ( line_stripped.isupper() or # Headers line_stripped.startswith(('•', '-', '*', '1.', '2.')) or # Lists any(keyword in line_stripped.lower() for keyword in ['summary', 'conclusion', 'important', 'key', 'main']) or len(line_stripped) < 100 # Short lines often important ) if is_important and current_length + len(line) < remaining_chars: important_lines.append(line) current_length += len(line) if important_lines: middle_section = '\n'.join(important_lines) # Combine sections if middle_section: return f"{beginning}\n\n[KEY SECTIONS]\n{middle_section}\n\n[ENDING]\n{ending}" else: return f"{beginning}\n\n[...content truncated...]\n\n{ending}" def _create_fallback_summary(self, content: str, filename: str) -> str: """Create intelligent fallback summary when LLM fails.""" # Extract first meaningful portion preview = content[:2000].strip() # Basic content analysis lines = content.split('\n') non_empty_lines = [line.strip() for line in lines if line.strip()] # Try to identify content type content_type = "document" if filename.endswith('.py'): content_type = "Python code" elif filename.endswith('.js'): content_type = "JavaScript code" elif filename.endswith('.md'): content_type = "Markdown document" elif filename.endswith('.txt'): content_type = "text document" elif filename.endswith('.csv'): content_type = "CSV data file" # Generate intelligent fallback return f"This {content_type} '{filename}' contains {len(content):,} characters across {len(non_empty_lines)} lines. Preview: {preview[:200]}{'...' if len(preview) > 200 else ''} This file would be useful for understanding the specific content and context it provides." def _extract_content(self, file_content: bytes, filename: str, mime_type: str) -> str: """Extract text content from file bytes.""" file_extension = Path(filename).suffix.lower() try: # Handle text-based files (including JSON, XML, CSV, etc.) if (file_extension in ['.txt', '.json', '.xml', '.csv', '.yml', '.yaml', '.md', '.log', '.ini', '.cfg', '.conf'] or mime_type.startswith('text/') or mime_type in ['application/json', 'application/xml', 'text/xml']): detected = chardet.detect(file_content) encoding = detected.get('encoding', 'utf-8') try: return file_content.decode(encoding) except UnicodeDecodeError: return file_content.decode('utf-8', errors='replace') elif file_extension == '.pdf': pdf_reader = PyPDF2.PdfReader(io.BytesIO(file_content)) return '\n\n'.join(page.extract_text() for page in pdf_reader.pages) elif file_extension == '.docx': doc = docx.Document(io.BytesIO(file_content)) return '\n'.join(paragraph.text for paragraph in doc.paragraphs) # For any other file type, try to decode as text (fallback) else: try: detected = chardet.detect(file_content) encoding = detected.get('encoding', 'utf-8') content = file_content.decode(encoding) # Only return if it seems to be mostly text content if len([c for c in content[:1000] if c.isprintable() or c.isspace()]) > 800: return content except: pass # If we can't extract text content, return a placeholder return f"[Binary file: {filename}] - Content cannot be extracted as text, but file is stored and available for download." except Exception as e: logger.error(f"Error extracting content from {filename}: {str(e)}") return f"[Error extracting content from {filename}] - File is stored but content extraction failed: {str(e)}"