import gradio as gr
import os
import json
import uuid
import asyncio
from datetime import datetime
from typing import List, Dict, Any, Optional, Generator
import logging
# Import required libraries
from huggingface_hub import InferenceClient
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.docstore.document import Document
# Import document parsers
import PyPDF2
from pptx import Presentation
import pandas as pd
from docx import Document as DocxDocument
import io
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Get HuggingFace token from environment
HF_TOKEN = os.getenv("hf_token")
if not HF_TOKEN:
raise ValueError("HuggingFace token not found in environment variables")
# Initialize HuggingFace Inference Client
client = InferenceClient(model="meta-llama/Llama-3.1-8B-Instruct", token=HF_TOKEN)
# Initialize embeddings
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
class MCPMessage:
"""Model Context Protocol Message Structure"""
def __init__(self, sender: str, receiver: str, msg_type: str,
trace_id: str = None, payload: Dict = None):
self.sender = sender
self.receiver = receiver
self.type = msg_type
self.trace_id = trace_id or str(uuid.uuid4())
self.payload = payload or {}
self.timestamp = datetime.now().isoformat()
def to_dict(self):
return {
"sender": self.sender,
"receiver": self.receiver,
"type": self.type,
"trace_id": self.trace_id,
"payload": self.payload,
"timestamp": self.timestamp
}
class MessageBus:
"""In-memory message bus for MCP communication"""
def __init__(self):
self.messages = []
self.subscribers = {}
def publish(self, message: MCPMessage):
"""Publish message to the bus"""
self.messages.append(message)
logger.info(f"Message published: {message.sender} -> {message.receiver} [{message.type}]")
# Notify subscribers
if message.receiver in self.subscribers:
for callback in self.subscribers[message.receiver]:
callback(message)
def subscribe(self, agent_name: str, callback):
"""Subscribe agent to receive messages"""
if agent_name not in self.subscribers:
self.subscribers[agent_name] = []
self.subscribers[agent_name].append(callback)
# Global message bus
message_bus = MessageBus()
class IngestionAgent:
"""Agent responsible for document parsing and preprocessing"""
def __init__(self, message_bus: MessageBus):
self.name = "IngestionAgent"
self.message_bus = message_bus
self.message_bus.subscribe(self.name, self.handle_message)
self.text_splitter = RecursiveCharacterTextSplitter(
chunk_size=1000,
chunk_overlap=200
)
def handle_message(self, message: MCPMessage):
"""Handle incoming MCP messages"""
if message.type == "INGESTION_REQUEST":
self.process_documents(message)
def parse_pdf(self, file_path: str) -> str:
"""Parse PDF document"""
try:
with open(file_path, 'rb') as file:
pdf_reader = PyPDF2.PdfReader(file)
text = ""
for page in pdf_reader.pages:
text += page.extract_text()
return text
except Exception as e:
logger.error(f"Error parsing PDF: {e}")
return ""
def parse_pptx(self, file_path: str) -> str:
"""Parse PPTX document"""
try:
prs = Presentation(file_path)
text = ""
for slide in prs.slides:
for shape in slide.shapes:
if hasattr(shape, "text"):
text += shape.text + "\n"
return text
except Exception as e:
logger.error(f"Error parsing PPTX: {e}")
return ""
def parse_csv(self, file_path: str) -> str:
"""Parse CSV document"""
try:
df = pd.read_csv(file_path)
return df.to_string()
except Exception as e:
logger.error(f"Error parsing CSV: {e}")
return ""
def parse_docx(self, file_path: str) -> str:
"""Parse DOCX document"""
try:
doc = DocxDocument(file_path)
text = ""
for paragraph in doc.paragraphs:
text += paragraph.text + "\n"
return text
except Exception as e:
logger.error(f"Error parsing DOCX: {e}")
return ""
def parse_txt(self, file_path: str) -> str:
"""Parse TXT/Markdown document"""
try:
with open(file_path, 'r', encoding='utf-8') as file:
return file.read()
except Exception as e:
logger.error(f"Error parsing TXT: {e}")
return ""
def process_documents(self, message: MCPMessage):
"""Process uploaded documents"""
files = message.payload.get("files", [])
processed_docs = []
for file_path in files:
file_ext = os.path.splitext(file_path)[1].lower()
# Parse document based on file type
if file_ext == '.pdf':
text = self.parse_pdf(file_path)
elif file_ext == '.pptx':
text = self.parse_pptx(file_path)
elif file_ext == '.csv':
text = self.parse_csv(file_path)
elif file_ext == '.docx':
text = self.parse_docx(file_path)
elif file_ext in ['.txt', '.md']:
text = self.parse_txt(file_path)
else:
logger.warning(f"Unsupported file type: {file_ext}")
continue
if text:
# Split text into chunks
chunks = self.text_splitter.split_text(text)
docs = [Document(page_content=chunk, metadata={"source": file_path})
for chunk in chunks]
processed_docs.extend(docs)
# Send processed documents to RetrievalAgent
response = MCPMessage(
sender=self.name,
receiver="RetrievalAgent",
msg_type="INGESTION_COMPLETE",
trace_id=message.trace_id,
payload={"documents": processed_docs}
)
self.message_bus.publish(response)
class RetrievalAgent:
"""Agent responsible for embedding and semantic retrieval"""
def __init__(self, message_bus: MessageBus):
self.name = "RetrievalAgent"
self.message_bus = message_bus
self.message_bus.subscribe(self.name, self.handle_message)
self.vector_store = None
def handle_message(self, message: MCPMessage):
"""Handle incoming MCP messages"""
if message.type == "INGESTION_COMPLETE":
self.create_vector_store(message)
elif message.type == "RETRIEVAL_REQUEST":
self.retrieve_context(message)
def create_vector_store(self, message: MCPMessage):
"""Create vector store from processed documents"""
documents = message.payload.get("documents", [])
if documents:
try:
self.vector_store = FAISS.from_documents(documents, embeddings)
logger.info(f"Vector store created with {len(documents)} documents")
# Notify completion
response = MCPMessage(
sender=self.name,
receiver="CoordinatorAgent",
msg_type="VECTORSTORE_READY",
trace_id=message.trace_id,
payload={"status": "ready"}
)
self.message_bus.publish(response)
except Exception as e:
logger.error(f"Error creating vector store: {e}")
def retrieve_context(self, message: MCPMessage):
"""Retrieve relevant context for a query"""
query = message.payload.get("query", "")
k = message.payload.get("k", 3)
if self.vector_store and query:
try:
docs = self.vector_store.similarity_search(query, k=k)
context = [{"content": doc.page_content, "source": doc.metadata.get("source", "")}
for doc in docs]
response = MCPMessage(
sender=self.name,
receiver="LLMResponseAgent",
msg_type="CONTEXT_RESPONSE",
trace_id=message.trace_id,
payload={
"query": query,
"retrieved_context": context,
"top_chunks": [doc.page_content for doc in docs]
}
)
self.message_bus.publish(response)
except Exception as e:
logger.error(f"Error retrieving context: {e}")
class LLMResponseAgent:
"""Agent responsible for generating LLM responses"""
def __init__(self, message_bus: MessageBus):
self.name = "LLMResponseAgent"
self.message_bus = message_bus
self.message_bus.subscribe(self.name, self.handle_message)
def handle_message(self, message: MCPMessage):
"""Handle incoming MCP messages"""
if message.type == "CONTEXT_RESPONSE":
self.generate_response(message)
def generate_response(self, message: MCPMessage):
"""Generate response using retrieved context"""
query = message.payload.get("query", "")
context = message.payload.get("retrieved_context", [])
# Build context string
context_text = "\n\n".join([f"Source: {ctx['source']}\nContent: {ctx['content']}"
for ctx in context])
# Create messages for conversational format
messages = [
{
"role": "system",
"content": "You are a helpful assistant. Based on the provided context below, answer the user's question accurately and comprehensively. Cite the sources if possible.",
},
{
"role": "user",
"content": f"Context:\n\n{context_text}\n\nQuestion: {query}"
}
]
try:
# Use client.chat_completion for conversational models
response_stream = client.chat_completion(
messages=messages,
max_tokens=512,
temperature=0.7,
stream=True
)
# Send streaming response
response = MCPMessage(
sender=self.name,
receiver="CoordinatorAgent",
msg_type="LLM_RESPONSE_STREAM",
trace_id=message.trace_id,
payload={
"query": query,
"response_stream": response_stream,
"context": context
}
)
self.message_bus.publish(response)
except Exception as e:
logger.error(f"Error generating response: {e}")
# Send an error stream back
error_msg = f"Error from LLM: {e}"
def error_generator():
yield error_msg
response = MCPMessage(
sender=self.name,
receiver="CoordinatorAgent",
msg_type="LLM_RESPONSE_STREAM",
trace_id=message.trace_id,
payload={"response_stream": error_generator()}
)
self.message_bus.publish(response)
class CoordinatorAgent:
"""Coordinator agent that orchestrates the entire workflow"""
def __init__(self, message_bus: MessageBus):
self.name = "CoordinatorAgent"
self.message_bus = message_bus
self.message_bus.subscribe(self.name, self.handle_message)
self.current_response_stream = None
self.vector_store_ready = False
def handle_message(self, message: MCPMessage):
"""Handle incoming MCP messages"""
if message.type == "VECTORSTORE_READY":
self.vector_store_ready = True
elif message.type == "LLM_RESPONSE_STREAM":
self.current_response_stream = message.payload.get("response_stream")
def process_files(self, files):
"""Process uploaded files"""
if not files:
return "No files uploaded."
file_paths = [file.name for file in files]
# Send ingestion request
message = MCPMessage(
sender=self.name,
receiver="IngestionAgent",
msg_type="INGESTION_REQUEST",
payload={"files": file_paths}
)
self.message_bus.publish(message)
return f"Processing {len(files)} files: {', '.join([os.path.basename(fp) for fp in file_paths])}"
def handle_query(self, query: str, history: List) -> Generator[str, None, None]:
"""Handle user query and return streaming response"""
if not self.vector_store_ready:
yield "Please upload and process documents first."
return
# Send retrieval request
message = MCPMessage(
sender=self.name,
receiver="RetrievalAgent",
msg_type="RETRIEVAL_REQUEST",
payload={"query": query}
)
self.message_bus.publish(message)
# Wait for response and stream
import time
timeout = 20 # seconds
start_time = time.time()
while not self.current_response_stream and (time.time() - start_time) < timeout:
time.sleep(0.1)
if self.current_response_stream:
try:
# Stream tokens directly
for chunk in self.current_response_stream:
# The token is in chunk.choices[0].delta.content for chat_completion
if hasattr(chunk, 'choices') and chunk.choices:
token = chunk.choices[0].delta.content
if token:
yield token
else:
# Fallback for different response format
if hasattr(chunk, 'token'):
yield chunk.token
elif isinstance(chunk, str):
yield chunk
except Exception as e:
yield f"Error streaming response: {e}"
finally:
self.current_response_stream = None # Reset for next query
else:
yield "Timeout: No response received from LLM agent."
# Initialize agents
ingestion_agent = IngestionAgent(message_bus)
retrieval_agent = RetrievalAgent(message_bus)
llm_response_agent = LLMResponseAgent(message_bus)
coordinator_agent = CoordinatorAgent(message_bus)
def create_interface():
"""Create ChatGPT-style Gradio interface"""
with gr.Blocks(
theme=gr.themes.Base(),
css="""
/* Dark theme styling */
.gradio-container {
background-color: #1a1a1a !important;
color: #ffffff !important;
height: 100vh !important;
max-width: none !important;
padding: 0 !important;
}
/* Main container */
.main-container {
display: flex;
flex-direction: column;
height: 100vh;
background: linear-gradient(135deg, #1a1a1a 0%, #2d2d2d 100%);
}
/* Header */
.header {
background: rgba(255, 193, 7, 0.1);
border-bottom: 1px solid rgba(255, 193, 7, 0.2);
padding: 1rem 2rem;
backdrop-filter: blur(10px);
}
.header h1 {
color: #ffc107;
margin: 0;
font-size: 1.5rem;
font-weight: 600;
}
.header p {
color: #cccccc;
margin: 0.25rem 0 0 0;
font-size: 0.9rem;
}
/* Chat area - REDUCED HEIGHT */
.chat-container {
flex: 1;
display: flex;
flex-direction: column;
max-width: 1000px;
margin: 0 auto;
width: 100%;
padding: 1rem;
height: calc(100vh - 200px) !important; /* Reduced height */
}
/* Chatbot styling - SMALLER */
.gradio-chatbot {
height: 300px !important; /* Reduced from 500px */
max-height: 300px !important;
background: transparent !important;
border: none !important;
margin-bottom: 1rem;
overflow-y: auto !important;
box-shadow: 0 0 12px rgba(255, 193, 7, 0.1);
}
/* Input area */
.input-area {
background: rgba(45, 45, 45, 0.6);
border-radius: 16px;
padding: 1rem;
border: 1px solid rgba(255, 193, 7, 0.2);
backdrop-filter: blur(10px);
position: sticky;
bottom: 0;
}
/* File upload */
.upload-area {
background: rgba(255, 193, 7, 0.05);
border: 2px dashed rgba(255, 193, 7, 0.3);
border-radius: 12px;
padding: 1rem;
margin-bottom: 1rem;
transition: all 0.3s ease;
}
/* Buttons - YELLOW SEND BUTTON */
.send-btn {
background: linear-gradient(135deg, #ffc107 0%, #ff8f00 100%) !important;
color: #000000 !important;
border: none !important;
border-radius: 8px !important;
font-weight: 600 !important;
min-height: 40px !important;
}
.primary-btn {
background: linear-gradient(135deg, #ffc107 0%, #ff8f00 100%) !important;
color: #000000 !important;
border: none !important;
border-radius: 8px !important;
font-weight: 600 !important;
}
/* Text inputs */
.gradio-textbox input, .gradio-textbox textarea {
background: rgba(45, 45, 45, 0.8) !important;
color: #ffffff !important;
border: 1px solid rgba(255, 193, 7, 0.2) !important;
border-radius: 8px !important;
}
/* Processing indicator */
.processing-indicator {
background: rgba(255, 193, 7, 0.1);
border: 1px solid rgba(255, 193, 7, 0.3);
border-radius: 8px;
padding: 0.75rem;
margin: 0.5rem 0;
color: #ffc107;
text-align: center;
}
/* Input row styling */
.input-row {
display: flex !important;
gap: 10px !important;
align-items: end !important;
}
/* Message input */
.message-input {
flex: 1 !important;
min-height: 40px !important;
}
""",
title="Agentic RAG Assistant"
) as iface:
# Header
with gr.Row():
with gr.Column():
gr.HTML("""
""")
# Main layout with sidebar and chat
with gr.Row():
# Left sidebar for file upload
with gr.Column(scale=1):
gr.Markdown("### đ Document Upload")
file_upload = gr.File(
file_count="multiple",
file_types=[".pdf", ".pptx", ".csv", ".docx", ".txt", ".md"],
label="Upload Documents",
elem_classes=["upload-area"]
)
processing_status = gr.HTML(visible=False)
process_btn = gr.Button(
"Process Documents",
variant="primary",
elem_classes=["primary-btn"]
)
# gr.Markdown("### âšī¸ Architecture")
# gr.Markdown("""
# **Multi-Agent System:**
# - đ **IngestionAgent**: Document parsing
# - đ **RetrievalAgent**: Semantic search
# - đ¤ **LLMAgent**: Response generation
# - đ¯ **CoordinatorAgent**: Workflow orchestration
# **Features:**
# - Streaming responses
# - Multi-format support
# - Context-aware answers
# """)
# Right side - Chat interface
with gr.Column(scale=2):
gr.Markdown("### đŦ Chat Interface")
# Chatbot with reduced height
chatbot = gr.Chatbot(
height=300, # Reduced height
elem_classes=["gradio-chatbot"],
show_copy_button=True,
type="messages",
placeholder="Upload documents first, then start chatting!"
)
# Input area with improved layout
with gr.Row(elem_classes=["input-row"]):
msg_input = gr.Textbox(
placeholder="Ask about your documents...",
label="Message",
scale=4,
elem_classes=["message-input"],
show_label=False,
autofocus=True
)
send_btn = gr.Button(
"Send",
scale=1,
elem_classes=["send-btn"],
size="sm"
)
# Examples
gr.Examples(
examples=[
"What are the main topics discussed?",
"Summarize the key findings",
"What metrics are mentioned?",
"What are the recommendations?"
],
inputs=msg_input,
label="Example Questions"
)
# State to track document processing
doc_processed = gr.State(False)
# Event handlers
def handle_file_upload_and_process(files):
if not files:
return gr.update(visible=False), False
# Show processing indicator
processing_html = f"""
đ Processing {len(files)} documents... Please wait.
"""
# Process files
try:
result = coordinator_agent.process_files(files)
# Wait a moment for processing to complete
import time
time.sleep(3)
success_html = """
Documents processed successfully! You can now ask questions.
"""
return gr.update(value=success_html, visible=True), True
except Exception as e:
error_html = f"""
â Error processing documents: {str(e)}
"""
return gr.update(value=error_html, visible=True), False
def respond(message, history, doc_ready):
if not doc_ready:
# Show error message
history.append({"role": "user", "content": message})
history.append({"role": "assistant", "content": " Please upload and process documents first."})
return history, ""
if not message.strip():
return history, message
# Add user message
history.append({"role": "user", "content": message})
history.append({"role": "assistant", "content": ""})
# Stream response
try:
for token in coordinator_agent.handle_query(message, history):
history[-1]["content"] += token
yield history, ""
except Exception as e:
history[-1]["content"] = f"â Error: {str(e)}"
yield history, ""
# Event bindings
process_btn.click(
handle_file_upload_and_process,
inputs=[file_upload],
outputs=[processing_status, doc_processed]
)
send_btn.click(
respond,
inputs=[msg_input, chatbot, doc_processed],
outputs=[chatbot, msg_input],
show_progress=True
)
msg_input.submit(
respond,
inputs=[msg_input, chatbot, doc_processed],
outputs=[chatbot, msg_input],
show_progress=True
)
return iface
# Launch the application
if __name__ == "__main__":
demo = create_interface()
demo.launch(
share=True,
server_name="0.0.0.0",
server_port=7860
)