Update: ChatGPT-style file upload, PDF/DOCX/PPT support

app_backup.py

@@ -0,0 +1,596 @@
+
+"""
+Krish Mind Local Server (GGUF)
+==============================
+Works with index_local.html
+Features: GGUF Model + Web Search + Image Generation + RAG + File Upload
+"""
+
+import os
+import sys
+import urllib.parse
+import pickle
+import tempfile
+from datetime import datetime
+
+print("=" * 60)
+print("🧠 Krish Mind Local Server (GGUF)")
+print("=" * 60)
+
+# --- Core dependencies ---
+try:
+    from llama_cpp import Llama
+    print("✅ llama-cpp-python")
+except ImportError:
+    print("❌ Run: pip install llama-cpp-python")
+    sys.exit(1)
+
+try:
+    from fastapi import FastAPI, UploadFile, File
+    from fastapi.middleware.cors import CORSMiddleware
+    from pydantic import BaseModel
+    import uvicorn
+    print("✅ fastapi + uvicorn")
+except ImportError:
+    print("❌ Run: pip install fastapi uvicorn python-multipart")
+    sys.exit(1)
+
+# --- Config ---
+GGUF_PATH = "d:/Krish Mind/gguf/krish-mind-standalone-Q4.gguf"
+EMBEDDINGS_FILE = "../data/krce_embeddings.pkl"
+DATA_FILE = "../data/krce_college_data.jsonl"
+
+# --- Load GGUF Model ---
+print(f"\n⏳ Loading GGUF model...")
+try:
+    model = Llama(
+        model_path=GGUF_PATH,
+        n_ctx=4096,
+        n_gpu_layers=0,
+        verbose=False
+    )
+    print("✅ Model loaded!")
+except Exception as e:
+    print(f"❌ Model error: {e}")
+    sys.exit(1)
+
+# --- DuckDuckGo Web Search ---
+print("\n📦 Loading optional features...")
+ddgs = None
+try:
+    import warnings
+    warnings.filterwarnings("ignore")
+    from duckduckgo_search import DDGS
+    ddgs = DDGS()
+    print("✅ DuckDuckGo web search")
+except Exception as e:
+    print(f"⚠️ Web search disabled: {e}")
+
+# --- RAG SETUP (Load Pre-computed Embeddings) ---
+print("📚 Loading Knowledge Base...")
+knowledge_base = []
+doc_embeddings = None
+rag_model = None
+
+# Try to load pre-computed embeddings first
+if os.path.exists(EMBEDDINGS_FILE):
+    try:
+        import numpy as np
+        from sentence_transformers import SentenceTransformer
+        
+        print(f"📂 Loading pre-computed embeddings from {EMBEDDINGS_FILE}...")
+        with open(EMBEDDINGS_FILE, 'rb') as f:
+            data = pickle.load(f)
+        
+        knowledge_base = data['knowledge_base']
+        doc_embeddings = data['embeddings']
+        
+        # Load the model for query encoding (needed for search)
+        rag_model = SentenceTransformer(data.get('model_name', 'all-MiniLM-L6-v2'))
+        print(f"✅ Embeddings loaded! ({len(knowledge_base)} facts)")
+        
+    except Exception as e:
+        print(f"⚠️ Could not load embeddings: {e}")
+        print("   Falling back to live embedding...")
+
+# Fallback: compute embeddings if pkl not found
+if doc_embeddings is None and os.path.exists(DATA_FILE):
+    try:
+        from sentence_transformers import SentenceTransformer
+        import numpy as np
+        import json
+        
+        rag_model = SentenceTransformer('all-MiniLM-L6-v2')
+        print("✅ Embedding model loaded (SentenceTransformer)")
+        
+        with open(DATA_FILE, 'r') as f:
+            for line in f:
+                if line.strip():
+                    try:
+                        knowledge_base.append(json.loads(line))
+                    except:
+                        pass
+        
+        if knowledge_base:
+            docs = [f"{k['instruction']} {k['output']}" for k in knowledge_base]
+            doc_embeddings = rag_model.encode(docs)
+            print(f"✅ Embeddings computed! ({len(knowledge_base)} facts)")
+            print("   ⚠️ Run 'python scripts/build_embeddings.py' for faster startup!")
+            
+    except Exception as e:
+        print(f"❌ RAG disabled: {e}")
+        rag_model = None
+else:
+    if doc_embeddings is None:
+        print("⚠️ Data file not found! RAG disabled.")
+
+# Initialize Cross-Encoder for re-ranking
+cross_encoder = None
+try:
+    from sentence_transformers import CrossEncoder
+    cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
+    print("✅ Cross-Encoder loaded for re-ranking")
+except Exception as e:
+    print(f"⚠️ Cross-Encoder not available: {e}")
+
+# ============================================
+# FILE UPLOAD RAG: Session-based document analysis
+# ============================================
+# Store uploaded file embeddings per session (in-memory)
+session_file_data = {}
+
+def extract_text_from_file(file_path: str, filename: str) -> str:
+    """Extract text from uploaded file"""
+    text = ""
+    ext = filename.lower().split('.')[-1]
+    
+    try:
+        if ext == 'txt':
+            with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
+                text = f.read()
+        
+        elif ext == 'pdf':
+            try:
+                import PyPDF2
+                with open(file_path, 'rb') as f:
+                    reader = PyPDF2.PdfReader(f)
+                    for page in reader.pages:
+                        text += page.extract_text() + "\n"
+            except ImportError:
+                try:
+                    import fitz  # PyMuPDF
+                    doc = fitz.open(file_path)
+                    for page in doc:
+                        text += page.get_text() + "\n"
+                    doc.close()
+                except ImportError:
+                    return "Error: Install PyPDF2 or PyMuPDF to read PDFs"
+        
+        elif ext in ['doc', 'docx']:
+            try:
+                import docx
+                doc = docx.Document(file_path)
+                text = "\n".join([para.text for para in doc.paragraphs])
+            except ImportError:
+                return "Error: Install python-docx to read Word files"
+        
+        elif ext in ['md', 'json', 'csv']:
+            with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
+                text = f.read()
+        
+        else:
+            # Try reading as plain text
+            with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
+                text = f.read()
+    
+    except Exception as e:
+        return f"Error reading file: {e}"
+    
+    return text.strip()
+
+def chunk_text(text: str, chunk_size: int = 500, overlap: int = 50) -> list:
+    """Split text into chunks for embedding"""
+    words = text.split()
+    chunks = []
+    
+    for i in range(0, len(words), chunk_size - overlap):
+        chunk = " ".join(words[i:i + chunk_size])
+        if chunk.strip():
+            chunks.append(chunk)
+    
+    return chunks
+
+# ============================================
+# ADVANCED RAG: Query Expansion + Hybrid Search
+# ============================================
+
+ABBREVIATIONS = {
+    "aids": "AI&DS Artificial Intelligence and Data Science",
+    "ai&ds": "AI&DS Artificial Intelligence and Data Science",
+    "aid": "AI&DS Artificial Intelligence and Data Science",
+    "aiml": "AI&ML Artificial Intelligence and Machine Learning",
+    "ai&ml": "AI&ML Artificial Intelligence and Machine Learning",
+    "cse": "Computer Science Engineering CSE",
+    "ece": "Electronics Communication Engineering ECE",
+    "eee": "Electrical Electronics Engineering EEE",
+    "mech": "Mechanical Engineering",
+    "it": "Information Technology IT",
+    "hod": "Head of Department HOD",
+    "mam": "madam professor female faculty",
+    "sir": "male professor faculty",
+    "staffs": "staff faculty members",
+    "sase": "Sasikumar Sasidevi",
+    "krce": "K. Ramakrishnan College of Engineering",
+}
+
+def expand_query(query):
+    """Expand abbreviations and add synonyms for better matching"""
+    expanded = query.lower()
+    for abbr, full in ABBREVIATIONS.items():
+        if abbr in expanded.split():
+            expanded = expanded + " " + full
+    return expanded
+
+def search_krce(query, threshold=0.25):
+    """Advanced RAG: Query Expansion + Vector Search + Cross-Encoder Re-ranking"""
+    if not rag_model or doc_embeddings is None:
+        return ""
+    
+    try:
+        expanded_query = expand_query(query)
+        
+        print(f"\n📊 RAG Search")
+        print(f"  Original: '{query}'")
+        print(f"  Expanded: '{expanded_query}'")
+        print("-" * 50)
+        
+        from sklearn.metrics.pairwise import cosine_similarity
+        q_emb = rag_model.encode([expanded_query])
+        vector_scores = cosine_similarity(q_emb, doc_embeddings).flatten()
+        
+        top_indices = vector_scores.argsort()[-10:][::-1]
+        top_candidates = [(idx, vector_scores[idx]) for idx in top_indices]
+
+        print("Vector Search Results:")
+        for i, (idx, v) in enumerate(top_candidates[:5]):
+            instruction = knowledge_base[idx]['instruction'][:35]
+            print(f"  #{i+1} V:{v:.3f} | {instruction}...")
+        
+        if cross_encoder:
+            pairs = [[query, f"{knowledge_base[idx]['instruction']} {knowledge_base[idx]['output']}"] 
+                     for idx, _ in top_candidates]
+            ce_scores = cross_encoder.predict(pairs)
+            
+            final_ranking = sorted(zip(top_candidates, ce_scores), key=lambda x: x[1], reverse=True)
+            
+            print("Cross-Encoder Re-ranking:")
+            for i, ((idx, v), ce) in enumerate(final_ranking[:3]):
+                instruction = knowledge_base[idx]['instruction'][:35]
+                print(f"  #{i+1} CE:{ce:.3f} | {instruction}...")
+            print("-" * 50)
+            
+            final_context = []
+            print(f"✅ RAG Retrieval (Top 5):")
+            for i, ((idx, v), ce) in enumerate(final_ranking[:5]):
+                if ce > -6.0:
+                    content = knowledge_base[idx]['output']
+                    final_context.append(content)
+                    print(f"  Took #{i+1}: {knowledge_base[idx]['instruction'][:30]}...")
+            
+            if final_context:
+                return "\n\n".join(final_context)
+        else:
+            final_context = []
+            for i, (idx, score) in enumerate(top_candidates[:5]):
+                if score > threshold:
+                    final_context.append(knowledge_base[idx]['output'])
+            
+            if final_context:
+                return "\n\n".join(final_context)
+        
+        print("❌ No confident match found")
+        return ""
+        
+    except Exception as e:
+        print(f"RAG Error: {e}")
+        return ""
+
+def search_file_context(query: str, session_id: str) -> str:
+    """Search uploaded file for relevant context"""
+    if session_id not in session_file_data or not rag_model:
+        print(f"⚠️ File context not found: session_id={session_id}, in_session={session_id in session_file_data}")
+        return ""
+    
+    # Maximum characters to return (prevent context overflow)
+    MAX_CONTEXT_CHARS = 2000
+    
+    try:
+        file_data = session_file_data[session_id]
+        chunks = file_data['chunks']
+        embeddings = file_data['embeddings']
+        filename = file_data.get('filename', 'uploaded file')
+        
+        # Detect general queries about the file (summarize, what's in the file, etc.)
+        general_triggers = ['summarize', 'summary', 'what is in', "what's in", 'tell me about', 
+                           'describe', 'overview', 'main points', 'key points', 'the file', 
+                           'the document', 'uploaded', 'attached', 'read the']
+        is_general_query = any(t in query.lower() for t in general_triggers)
+        
+        if is_general_query:
+            # For general queries, return content up to limit
+            print(f"📄 General file query detected - returning limited content")
+            all_content = ""
+            for chunk in chunks:
+                if len(all_content) + len(chunk) > MAX_CONTEXT_CHARS:
+                    break
+                all_content += chunk + "\n\n"
+            
+            if len(all_content) < sum(len(c) for c in chunks):
+                all_content += f"\n[...content truncated, showing {len(all_content)} of {sum(len(c) for c in chunks)} chars]"
+            
+            return f"[Content from {filename}]:\n\n{all_content.strip()}"
+        
+        # For specific queries, use semantic search
+        from sklearn.metrics.pairwise import cosine_similarity
+        q_emb = rag_model.encode([query])
+        scores = cosine_similarity(q_emb, embeddings).flatten()
+        
+        # Get top 3 most relevant chunks
+        top_indices = scores.argsort()[-3:][::-1]
+        context_parts = []
+        total_chars = 0
+        
+        for idx in top_indices:
+            if scores[idx] > 0.15 and total_chars < MAX_CONTEXT_CHARS:
+                chunk = chunks[idx]
+                if total_chars + len(chunk) > MAX_CONTEXT_CHARS:
+                    # Truncate to fit
+                    remaining = MAX_CONTEXT_CHARS - total_chars
+                    chunk = chunk[:remaining] + "..."
+                context_parts.append(chunk)
+                total_chars += len(chunk)
+        
+        if context_parts:
+            print(f"📄 File context found ({len(context_parts)} chunks, {total_chars} chars)")
+            return f"[Content from {filename}]:\n\n" + "\n\n".join(context_parts)
+        
+        # Fallback: if no good matches, return first chunk truncated
+        print(f"📄 Low confidence match - returning truncated first chunk")
+        first_chunk = chunks[0][:MAX_CONTEXT_CHARS] if chunks else ""
+        return f"[Content from {filename}]:\n\n{first_chunk}"
+        
+    except Exception as e:
+        print(f"File search error: {e}")
+        return ""
+
+def search_web(query):
+    if not ddgs:
+        return ""
+    try:
+        results = ddgs.text(query, max_results=3)
+        if not results:
+            return ""
+        return "\n\n".join([f"**{r['title']}**\n{r['body']}" for r in results])
+    except:
+        return ""
+
+# --- FastAPI ---
+app = FastAPI()
+app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
+
+class ChatRequest(BaseModel):
+    message: str
+    max_tokens: int = 512
+    temperature: float = 0.7
+    summary: str = ""
+    history: list = []
+    session_id: str = ""  # For file upload sessions
+
+class SummarizeRequest(BaseModel):
+    messages: list
+
+@app.get("/")
+async def root():
+    return {"name": "Krish Mind", "status": "online", "rag": rag_model is not None, "web": ddgs is not None}
+
+@app.post("/upload")
+async def upload_file(file: UploadFile = File(...), session_id: str = "default"):
+    """Upload a file for RAG analysis"""
+    if not rag_model:
+        return {"success": False, "error": "Embedding model not available"}
+    
+    try:
+        # Save uploaded file temporarily
+        suffix = '.' + file.filename.split('.')[-1] if '.' in file.filename else '.txt'
+        with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
+            content = await file.read()
+            tmp.write(content)
+            tmp_path = tmp.name
+        
+        # Extract text
+        print(f"📤 Processing uploaded file: {file.filename}")
+        text = extract_text_from_file(tmp_path, file.filename)
+        
+        # Clean up temp file
+        os.unlink(tmp_path)
+        
+        if text.startswith("Error"):
+            return {"success": False, "error": text}
+        
+        if not text.strip():
+            return {"success": False, "error": "Could not extract text from file"}
+        
+        # Chunk text
+        chunks = chunk_text(text)
+        if not chunks:
+            return {"success": False, "error": "File too small or empty"}
+        
+        # Create embeddings
+        print(f"🔄 Creating embeddings for {len(chunks)} chunks...")
+        embeddings = rag_model.encode(chunks)
+        
+        # Store in session
+        session_file_data[session_id] = {
+            "filename": file.filename,
+            "chunks": chunks,
+            "embeddings": embeddings,
+            "full_text": text[:2000]  # First 2000 chars for context
+        }
+        
+        print(f"✅ File processed: {len(chunks)} chunks, {len(text)} chars")
+        
+        return {
+            "success": True,
+            "filename": file.filename,
+            "chunks": len(chunks),
+            "chars": len(text),
+            "preview": text[:200] + "..." if len(text) > 200 else text
+        }
+        
+    except Exception as e:
+        print(f"❌ Upload error: {e}")
+        return {"success": False, "error": str(e)}
+
+@app.delete("/upload/{session_id}")
+async def clear_file(session_id: str):
+    """Clear uploaded file from session"""
+    if session_id in session_file_data:
+        del session_file_data[session_id]
+        return {"success": True, "message": "File cleared"}
+    return {"success": False, "message": "No file found for session"}
+
+@app.post("/summarize")
+async def summarize(request: SummarizeRequest):
+    """Summarize older messages to compress context"""
+    try:
+        messages_text = ""
+        for msg in request.messages:
+            role = msg.get("role", "user")
+            content = msg.get("content", "")
+            messages_text += f"{role.capitalize()}: {content}\n"
+        
+        summary_prompt = f"""<|start_header_id|>system<|end_header_id|>
+
+You are a conversation summarizer. Condense the following conversation into a brief summary (2-3 sentences max) that captures the key topics and context. Focus on what was discussed, not exact words.<|eot_id|><|start_header_id|>user<|end_header_id|>
+
+Summarize this conversation:
+{messages_text}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
+
+Summary: """
+        
+        output = model(summary_prompt, max_tokens=150, temperature=0.3, stop=["<|eot_id|>"], echo=False)
+        summary = output["choices"][0]["text"].strip()
+        print(f"📝 Summarized {len(request.messages)} messages: {summary[:50]}...")
+        return {"summary": summary}
+    except Exception as e:
+        print(f"❌ Summarization error: {e}")
+        return {"summary": "", "error": str(e)}
+
+
+@app.post("/chat")
+async def chat(request: ChatRequest):
+    user_input = request.message
+    session_id = request.session_id or "default"
+    
+    # Image generation
+    img_triggers = ["generate image", "create image", "draw", "imagine"]
+    if any(t in user_input.lower() for t in img_triggers):
+        prompt = user_input
+        for t in img_triggers:
+            prompt = prompt.lower().replace(t, "")
+        prompt = prompt.strip()
+        if prompt:
+            url = f"https://image.pollinations.ai/prompt/{urllib.parse.quote(prompt)}"
+            return {"response": f"Here's your image of **{prompt}**:\n\n![{prompt}]({url})"}
+    
+    # RAG Search (college knowledge base)
+    rag_context = ""
+    if rag_model:
+        rag_context = search_krce(user_input)
+        if rag_context:
+            print(f"🧠 RAG Context found: {rag_context[:50]}...")
+    
+    # File context (uploaded document)
+    file_context = ""
+    if session_id in session_file_data:
+        file_context = search_file_context(user_input, session_id)
+        if file_context:
+            print(f"📄 File Context found: {file_context[:50]}...")
+
+    # Web search
+    web_context = ""
+    search_triggers = ["search", "find", "latest", "news", "who is", "what is", "when", "where", "how"]
+    if ddgs and any(t in user_input.lower() for t in search_triggers):
+        if len(user_input.split()) > 2:
+            print(f"🔎 Searching web...")
+            web_context = search_web(user_input)
+    
+    # Build prompt
+    now = datetime.now().strftime("%A, %B %d, %Y %I:%M %p")
+    
+    sys_prompt = f"""You are Krish Mind, a helpful AI assistant created by Krish CS. Current time: {now}
+
+IMPORTANT STRICT RULES:
+1. IDENTITY: You were created by Krish CS. Do NOT claim to be created by anyone mentioned in the context (like faculty, HODs, or staff). If the context mentions a name, that person is a subject of the data, NOT your creator.
+2. CONTEXT USAGE: Use the provided context to answer questions. If the context contains a list (e.g., faculty names), make sure to include ALL items found in the context chunks.
+3. FORMATTING: Use Markdown. For letters, use **bold** for headers (e.g., **Subject:**) and use DOUBLE LINE BREAKS between sections (Place, Date, From, To, Subject, Body) to create clear distinct paragraphs.
+4. AMBIGUITY: 'AID' or 'AIDS' in this context ALWAYS refers to 'Artificial Intelligence and Data Science', NEVER 'Aerospace' or 'Disease'.
+5. ACCURACY: If the context contains a name like 'Mrs. C. Rani', she is a faculty member. Do NOT say "I was created by Mrs. C. Rani".
+"""
+    
+    if file_context:
+        sys_prompt += f"\n\nUploaded Document Context:\n{file_context}"
+    
+    if rag_context:
+        sys_prompt += f"\n\nKnowledge Base Context:\n{rag_context}"
+        
+    if web_context:
+        sys_prompt += f"\n\nWeb Results:\n{web_context}"
+    
+    if request.summary:
+        sys_prompt += f"\n\nPrevious conversation summary:\n{request.summary}"
+    
+    # Build history context
+    history_context = ""
+    if request.history:
+        for msg in request.history[-6:]:
+            role = msg.get("role", "user")
+            content = msg.get("content", "")
+            if role == "user":
+                history_context += f"<|start_header_id|>user<|end_header_id|>\n\n{content}<|eot_id|>"
+            else:
+                history_context += f"<|start_header_id|>assistant<|end_header_id|>\n\n{content}<|eot_id|>"
+    
+    # Build full prompt
+    if history_context:
+        full_prompt = f"""<|start_header_id|>system<|end_header_id|>
+
+{sys_prompt}<|eot_id|>{history_context}<|start_header_id|>user<|end_header_id|>
+
+{user_input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
+
+"""
+    else:
+        full_prompt = f"""<|start_header_id|>system<|end_header_id|>
+
+{sys_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>
+
+{user_input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
+
+"""
+    
+    try:
+        print(f"💬 Generating response...")
+        output = model(full_prompt, max_tokens=request.max_tokens, temperature=request.temperature, stop=["<|eot_id|>"], echo=False)
+        response = output["choices"][0]["text"].strip()
+        print(f"✅ Done")
+        return {"response": response}
+    except Exception as e:
+        return {"response": f"Error: {e}"}
+
+if __name__ == "__main__":
+    print("\n" + "=" * 60)
+    print("🚀 Server running at: http://127.0.0.1:8000")
+    print("📱 Open index_local.html in your browser")
+    print("=" * 60 + "\n")
+    uvicorn.run(app, host="0.0.0.0", port=8000)

requirements.txt

@@ -1,3 +1,4 @@
+# Krish Mind Deployment Requirements
 fastapi==0.109.0
 uvicorn==0.27.0
 pydantic==2.6.0
@@ -7,3 +8,12 @@ scikit-learn
 duckduckgo-search>=5.0
 python-multipart
 huggingface_hub>=0.20.0
+
+# For PDF file support
+PyPDF2
+
+# For Word document support
+python-docx
+
+# For PowerPoint support
+python-pptx

static/index.html

@@ -990,6 +990,213 @@
             height: 16px;
         }
 
+        /* ============================================ */
+        /* FILE CARD - ChatGPT Style */
+        /* ============================================ */
+
+        .file-card {
+            display: none;
+            align-items: center;
+            gap: 12px;
+            padding: 12px 14px;
+            background: var(--bg-tertiary);
+            border: 1px solid var(--border);
+            border-radius: 12px;
+            animation: fadeIn 0.3s ease;
+            position: relative;
+            max-width: 280px;
+        }
+
+        .file-card.show {
+            display: flex;
+        }
+
+        .file-card .file-icon-wrapper {
+            width: 40px;
+            height: 40px;
+            background: #2d8cbe;
+            border-radius: 8px;
+            display: flex;
+            align-items: center;
+            justify-content: center;
+            flex-shrink: 0;
+        }
+
+        .file-card .file-icon-wrapper svg {
+            width: 20px;
+            height: 20px;
+            stroke: white;
+        }
+
+        .file-card .file-info {
+            display: flex;
+            flex-direction: column;
+            gap: 2px;
+            min-width: 0;
+            flex: 1;
+        }
+
+        .file-card .file-name {
+            font-size: 14px;
+            font-weight: 500;
+            color: var(--text-primary);
+            overflow: hidden;
+            text-overflow: ellipsis;
+            white-space: nowrap;
+        }
+
+        .file-card .file-type {
+            font-size: 12px;
+            color: var(--text-muted);
+        }
+
+        .file-card .remove-file {
+            position: absolute;
+            top: -6px;
+            right: -6px;
+            width: 20px;
+            height: 20px;
+            background: var(--bg-secondary);
+            border: 1px solid var(--border);
+            border-radius: 50%;
+            cursor: pointer;
+            display: flex;
+            align-items: center;
+            justify-content: center;
+            font-size: 14px;
+            color: var(--text-muted);
+            transition: all 0.2s;
+        }
+
+        .file-card .remove-file:hover {
+            background: var(--bg-hover);
+            color: var(--text-primary);
+        }
+
+        #fileInput {
+            display: none;
+        }
+
+        /* File card in message bubble */
+        .message-file-card {
+            display: flex;
+            align-items: center;
+            gap: 10px;
+            padding: 10px 12px;
+            background: rgba(45, 140, 190, 0.15);
+            border: 1px solid rgba(45, 140, 190, 0.3);
+            border-radius: 10px;
+            margin-bottom: 8px;
+            max-width: 250px;
+        }
+
+        .message.user .message-file-card {
+            background: rgba(255, 255, 255, 0.15);
+            border-color: rgba(255, 255, 255, 0.2);
+        }
+
+        .message-file-card .file-icon-wrapper {
+            width: 32px;
+            height: 32px;
+            background: #2d8cbe;
+            border-radius: 6px;
+            display: flex;
+            align-items: center;
+            justify-content: center;
+            flex-shrink: 0;
+        }
+
+        .message-file-card .file-icon-wrapper svg {
+            width: 16px;
+            height: 16px;
+            stroke: white;
+        }
+
+        .message-file-card .file-info {
+            display: flex;
+            flex-direction: column;
+            gap: 1px;
+            min-width: 0;
+        }
+
+        .message-file-card .file-name {
+            font-size: 13px;
+            font-weight: 500;
+            overflow: hidden;
+            text-overflow: ellipsis;
+            white-space: nowrap;
+        }
+
+        .message-file-card .file-type {
+            font-size: 11px;
+            opacity: 0.7;
+        }
+
+        /* Input wrapper */
+        .input-wrapper {
+            display: flex;
+            align-items: flex-end;
+            gap: 8px;
+            padding: 12px 14px;
+            background: var(--bg-input);
+            border: 1px solid var(--border);
+            border-radius: 24px;
+            box-shadow: var(--shadow);
+            transition: border-color 0.2s, box-shadow 0.2s;
+        }
+
+        .input-wrapper:focus-within {
+            border-color: var(--accent);
+            box-shadow: 0 0 0 2px rgba(45, 140, 190, 0.2);
+        }
+
+        .input-content {
+            display: flex;
+            flex-direction: column;
+            flex: 1;
+            gap: 10px;
+            min-width: 0;
+        }
+
+        .input-row {
+            display: flex;
+            align-items: center;
+            gap: 8px;
+        }
+
+        .attach-btn {
+            width: 28px;
+            height: 28px;
+            display: flex;
+            align-items: center;
+            justify-content: center;
+            background: transparent;
+            border: none;
+            border-radius: 6px;
+            color: var(--accent);
+            cursor: pointer;
+            transition: all 0.2s;
+            flex-shrink: 0;
+        }
+
+        .attach-btn:hover {
+            background: rgba(45, 140, 190, 0.15);
+            color: var(--accent);
+            transform: scale(1.1);
+        }
+
+        .attach-btn svg {
+            width: 20px;
+            height: 20px;
+            stroke-width: 2.5;
+        }
+
+        @keyframes spin {
+            to {
+                transform: rotate(360deg);
+            }
+        }
+
         /* Copy Feedback */
         .action-btn.copied {
             color: #10b981;
@@ -1427,10 +1634,50 @@
                         </svg>
                         Create Image
                     </button>
+                    <button class="tool-btn" id="fileUploadBtn" onclick="document.getElementById('fileInput').click()">
+                        <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                            <path
+                                d="M21.44 11.05l-9.19 9.19a6 6 0 0 1-8.49-8.49l9.19-9.19a4 4 0 0 1 5.66 5.66l-9.2 9.19a2 2 0 0 1-2.83-2.83l8.49-8.48">
+                            </path>
+                        </svg>
+                        Attach
+                    </button>
                 </div>
+
                 <div class="input-wrapper">
-                    <textarea id="messageInput" placeholder="Message Krish Mind..." rows="1"
-                        onkeydown="handleKeyDown(event)" oninput="autoResize(this)"></textarea>
+                    <input type="file" id="fileInput" accept=".pdf,.txt,.doc,.docx,.ppt,.pptx,.md,.json,.csv,.xlsx,.xls"
+                        onchange="handleFileSelect(event)">
+
+                    <div class="input-content">
+                        <!-- File Card (ChatGPT Style) -->
+                        <div class="file-card" id="fileCard">
+                            <div class="file-icon-wrapper">
+                                <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <path d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8z"></path>
+                                    <polyline points="14 2 14 8 20 8"></polyline>
+                                </svg>
+                            </div>
+                            <div class="file-info">
+                                <span class="file-name" id="fileName"></span>
+                                <span class="file-type">Document</span>
+                            </div>
+                            <button class="remove-file" onclick="clearStagedFile()">×</button>
+                        </div>
+
+                        <!-- Input Row with + button and textarea -->
+                        <div class="input-row">
+                            <button class="attach-btn" onclick="document.getElementById('fileInput').click()"
+                                title="Attach file">
+                                <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <line x1="12" y1="5" x2="12" y2="19"></line>
+                                    <line x1="5" y1="12" x2="19" y2="12"></line>
+                                </svg>
+                            </button>
+                            <textarea id="messageInput" placeholder="Message Krish Mind..." rows="1"
+                                onkeydown="handleKeyDown(event)" oninput="autoResize(this)"></textarea>
+                        </div>
+                    </div>
+
                     <button class="send-btn" id="sendBtn" onclick="sendMessage()" disabled>
                         <svg width="16" height="16" viewBox="0 0 24 24" fill="white">
                             <path d="M2.01 21L23 12 2.01 3 2 10l15 2-15 2z" />
@@ -1513,18 +1760,22 @@
 
     <script>
         // State
-        let serverUrl = ''; // Empty = relative URL (works on HF Spaces or any host)
+        let serverUrl = ''; // Empty = relative URL (works on HF Spaces)
         let isConnected = true; // Always connected when hosted on same server
         let isGenerating = false; // Prevent multiple requests
         let imageGenMode = false; // Image generation tool toggle
         let messages = [];  // Current display messages
-        let chatHistory = JSON.parse(sessionStorage.getItem('krishMindChatHistory') || '[]');
+        let chatHistory = [];  // Chat history (clears on refresh)
         let currentChatId = null;
         let currentAbortController = null;  // For aborting ongoing requests
 
-        // Smart Memory State
-        let fullMessages = JSON.parse(sessionStorage.getItem('krishMindFullMessages') || '[]');  // Full conversation
-        let conversationSummary = sessionStorage.getItem('krishMindSummary') || '';  // Compressed context
+        // File Upload State
+        let sessionId = 'session_' + Date.now() + '_' + Math.random().toString(36).substr(2, 9);
+        let uploadedFile = null;  // Currently uploaded file info
+
+        // Smart Memory State (clears on refresh)
+        let fullMessages = [];  // Full conversation
+        let conversationSummary = '';  // Compressed context
         const SUMMARIZE_THRESHOLD = 10;  // Trigger summarization after this many messages
         const KEEP_RECENT = 6;  // Keep this many recent messages in full context
 
@@ -1709,8 +1960,22 @@
             const sendBtn = document.getElementById('sendBtn');
             const message = input.value.trim();
 
-            if (!message) return;
-            if (isGenerating) return; // Block if already generating
+            if (!message && !stagedFile) return;
+
+            // If already generating, this is a STOP request
+            if (isGenerating) {
+                console.log('⛔ Stopping generation...');
+                if (currentAbortController) {
+                    currentAbortController.abort();
+                    currentAbortController = null;
+                }
+                hideThinking();
+                isGenerating = false;
+                sendBtn.innerHTML = SEND_ICON;
+                sendBtn.disabled = !input.value.trim();
+                return;
+            }
+
             if (!isConnected) {
                 showConnectModal();
                 return;
@@ -1729,8 +1994,14 @@
             input.value = '';
             input.style.height = 'auto';
 
-            // Add user message
-            addMessage('user', message);
+            // Add user message (with file card if file attached)
+            const attachedFileName = stagedFileName;  // Capture before clearing
+            const fileToUpload = stagedFile;  // Store file reference before clearing UI
+            addMessage('user', message, attachedFileName);
+
+            // Clear file card immediately (no delay) - only clears UI, not file reference
+            const fileCard = document.getElementById('fileCard');
+            fileCard.classList.remove('show');
 
             // Scroll to bottom after adding user message
             const chatContainer = document.getElementById('chatContainer');
@@ -1771,9 +2042,51 @@
                 isGenerating = false;
                 sendBtn.innerHTML = SEND_ICON;
                 sendBtn.disabled = true;
+                stagedFile = null;
+                stagedFileName = null;
                 return;
             }
 
+            // Upload file if one was attached
+            const hadFile = !!fileToUpload;
+            if (hadFile) {
+                console.log(`📤 Uploading file: ${attachedFileName}`);
+                const formData = new FormData();
+                formData.append('file', fileToUpload);
+
+                try {
+                    const response = await fetch(`${serverUrl}/upload?session_id=${sessionId}`, {
+                        method: 'POST',
+                        body: formData
+                    });
+                    const data = await response.json();
+                    if (data.success) {
+                        console.log(`✅ File uploaded: ${data.filename} (${data.chunks} chunks)`);
+                    } else {
+                        console.error('Upload failed:', data.error);
+                        addMessage('assistant', 'Failed to process file: ' + data.error);
+                        isGenerating = false;
+                        sendBtn.innerHTML = SEND_ICON;
+                        sendBtn.disabled = true;
+                        stagedFile = null;
+                        stagedFileName = null;
+                        return;
+                    }
+                } catch (err) {
+                    console.error('Upload error:', err);
+                    addMessage('assistant', 'Failed to upload file. Please try again.');
+                    isGenerating = false;
+                    sendBtn.innerHTML = SEND_ICON;
+                    sendBtn.disabled = true;
+                    stagedFile = null;
+                    stagedFileName = null;
+                    return;
+                }
+            }
+            // Reset stagedFile reference
+            stagedFile = null;
+            stagedFileName = null;
+
             // Show thinking
             showThinking();
 
@@ -1818,7 +2131,8 @@
                         max_tokens: 1024,
                         temperature: 0.7,
                         summary: conversationSummary,
-                        history: recentHistory
+                        history: recentHistory,
+                        session_id: sessionId
                     }),
                     signal: currentAbortController.signal
                 });
@@ -1846,11 +2160,16 @@
                 isGenerating = false;
                 sendBtn.innerHTML = SEND_ICON;
                 sendBtn.disabled = true;
+
+                // Clear server file if one was uploaded (ChatGPT style - one-time use)
+                if (hadFile) {
+                    await clearServerFile();
+                }
             }
         }
 
         // Add message with action buttons
-        function addMessage(role, content) {
+        function addMessage(role, content, attachedFile = null) {
             const container = document.getElementById('chatMessages');
             const msgIndex = messages.length;
 
@@ -1863,6 +2182,25 @@
             });
             html = html.replace(/<\/code><\/pre>/g, '</code></pre></div>');
 
+            // File card HTML for user messages (ChatGPT style)
+            let fileCardHtml = '';
+            if (attachedFile && role === 'user') {
+                fileCardHtml = `
+                    <div class="message-file-card">
+                        <div class="file-icon-wrapper">
+                            <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                <path d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8z"></path>
+                                <polyline points="14 2 14 8 20 8"></polyline>
+                            </svg>
+                        </div>
+                        <div class="file-info">
+                            <span class="file-name">${attachedFile}</span>
+                            <span class="file-type">Document</span>
+                        </div>
+                    </div>
+                `;
+            }
+
             // Action buttons based on role
             const editIcon = '<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><path d="M11 4H4a2 2 0 0 0-2 2v14a2 2 0 0 0 2 2h14a2 2 0 0 0 2-2v-7"></path><path d="M18.5 2.5a2.121 2.121 0 0 1 3 3L12 15l-4 1 1-4 9.5-9.5z"></path></svg>';
 
@@ -1887,6 +2225,7 @@
             div.setAttribute('data-index', msgIndex);
             div.innerHTML = `
                 <div class="message-wrapper">
+                    ${fileCardHtml}
                     <div class="message-content">${html}</div>
                     ${actionsHtml}
                 </div>
@@ -1901,11 +2240,10 @@
             // Highlight code
             div.querySelectorAll('pre code').forEach(block => hljs.highlightElement(block));
 
-            messages.push({ role, content });
+            messages.push({ role, content, attachedFile });
 
-            // Smart Memory: Save to fullMessages and persist
+            // Smart Memory: Save to fullMessages (in-memory only, clears on refresh)
             fullMessages.push({ role, content });
-            sessionStorage.setItem('krishMindFullMessages', JSON.stringify(fullMessages));
         }
 
         // Copy message content with tick feedback
@@ -2238,6 +2576,88 @@
         document.getElementById('connectModal').addEventListener('click', (e) => {
             if (e.target.id === 'connectModal') hideConnectModal();
         });
+
+        // ==========================================
+        // FILE UPLOAD FUNCTIONS (ChatGPT Style)
+        // ==========================================
+
+        let stagedFile = null;  // File staged for upload with next message
+        let stagedFileName = null;  // Store filename for message display
+
+        function handleFileSelect(event) {
+            const file = event.target.files[0];
+            if (!file) return;
+
+            // Stage the file (don't upload yet)
+            stagedFile = file;
+            stagedFileName = file.name;
+
+            // Show file card in input area
+            const fileCard = document.getElementById('fileCard');
+            const fileNameEl = document.getElementById('fileName');
+            fileNameEl.textContent = file.name;
+            fileCard.classList.add('show');
+
+            // Enable send button since we have a file
+            document.getElementById('sendBtn').disabled = false;
+
+            console.log(`📎 File staged: ${file.name}`);
+
+            // Clear input so same file can be re-selected
+            event.target.value = '';
+        }
+
+        function clearStagedFile() {
+            stagedFile = null;
+            stagedFileName = null;
+            const fileCard = document.getElementById('fileCard');
+            fileCard.classList.remove('show');
+
+            // Disable send button if no text
+            const input = document.getElementById('messageInput');
+            document.getElementById('sendBtn').disabled = !input.value.trim();
+
+            console.log('📎 Staged file cleared');
+        }
+
+        async function uploadStagedFile() {
+            if (!stagedFile) return true;  // No file to upload
+
+            try {
+                const formData = new FormData();
+                formData.append('file', stagedFile);
+
+                console.log(`📤 Uploading ${stagedFile.name}...`);
+
+                const response = await fetch(`${serverUrl}/upload?session_id=${sessionId}`, {
+                    method: 'POST',
+                    body: formData
+                });
+
+                const data = await response.json();
+
+                if (data.success) {
+                    console.log(`✅ File uploaded: ${data.filename} (${data.chunks} chunks)`);
+                    return true;
+                } else {
+                    console.error('Upload failed:', data.error);
+                    return false;
+                }
+            } catch (err) {
+                console.error('Upload error:', err);
+                return false;
+            }
+        }
+
+        async function clearServerFile() {
+            try {
+                await fetch(`${serverUrl}/upload/${sessionId}`, {
+                    method: 'DELETE'
+                });
+            } catch (err) {
+                // Ignore errors when clearing
+            }
+        }
     </script>
 </body>