Adding program files

agents.py

@@ -0,0 +1,301 @@
+import os
+from typing import List, Dict, Any, Optional
+import google.generativeai as genai
+from openai import OpenAI
+import logging
+import json
+
+logger = logging.getLogger(__name__)
+
+class BaseAgent:
+    def __init__(self, use_gemini: bool = True, api_key: Optional[str] = None, 
+                 openrouter_model: Optional[str] = None, gemini_model: Optional[str] = None):
+        self.use_gemini = use_gemini
+        if use_gemini:
+            if not api_key:
+                raise ValueError("Gemini API key is required when use_gemini=True")
+            genai.configure(api_key=api_key)
+            self.gemini_model = gemini_model or "gemini-1.5-pro"  # Use a good default model
+        else:
+            self.openrouter_client = OpenAI(
+                base_url="https://openrouter.ai/api/v1",
+                api_key=api_key
+            )
+            self.model = openrouter_model or "anthropic/claude-3-opus:beta"
+
+    def _generate_with_gemini(self, prompt: str, system_prompt: str) -> str:
+        try:
+            model = genai.GenerativeModel(model_name=self.gemini_model)
+            # Combine system prompt and user prompt for Gemini
+            combined_prompt = f"System: {system_prompt}\n\nUser: {prompt}"
+            response = model.generate_content(
+                combined_prompt,
+                generation_config=genai.types.GenerationConfig(
+                    temperature=0.1
+                )
+            )
+            return response.text
+        except Exception as e:
+            logger.error(f"Gemini generation failed: {str(e)}")
+            raise
+
+    def _generate_with_openrouter(self, prompt: str, system_prompt: str) -> str:
+        completion = self.openrouter_client.chat.completions.create(
+            model=self.model,
+            messages=[
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": prompt}
+            ],
+            temperature=0.1,
+        )
+        return completion.choices[0].message.content
+
+    def generate(self, prompt: str, system_prompt: str) -> str:
+        try:
+            if self.use_gemini:
+                return self._generate_with_gemini(prompt, system_prompt)
+            else:
+                return self._generate_with_openrouter(prompt, system_prompt)
+        except Exception as e:
+            logger.error(f"Generation failed: {str(e)}")
+            raise
+
+class OrchestratorAgent(BaseAgent):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.system_prompt = """You are an expert research planner that develops comprehensive research strategies.
+        Your role is to create structured research plans that identify what information is needed and why.
+        Focus on the logical flow of information needed to answer the query comprehensively."""
+
+    def create_research_plan(self, query: str) -> Dict[str, List[str]]:
+        """Create a structured research plan with clear objectives"""
+        prompt = f"""Create a detailed research plan for the following query: {query}
+
+        Return a JSON object with the following structure:
+        {{
+            "core_concepts": ["list of fundamental concepts that need to be understood"],
+            "key_questions": ["specific questions that need to be answered"],
+            "information_requirements": ["specific pieces of information needed to answer each question"],
+            "research_priorities": ["ordered list of research priorities"]
+        }}
+
+        Make sure the plan flows logically and each item contributes to answering the main query."""
+        
+        response = self.generate(prompt, self.system_prompt)
+        try:
+            # Clean the response of any markdown formatting
+            cleaned_response = response.strip().replace('```json', '').replace('```', '').strip()
+            plan = json.loads(cleaned_response)
+            logger.info(f"Generated research plan: {json.dumps(plan, indent=2)}")
+            return plan
+        except:
+            logger.error(f"Failed to parse research plan: {response}")
+            # Return a basic plan structure if parsing fails
+            return {
+                "core_concepts": [query],
+                "key_questions": [query],
+                "information_requirements": [query],
+                "research_priorities": [query]
+            }
+
+    def evaluate_research_progress(self, plan: Dict[str, List[str]], gathered_info: List[str]) -> Dict[str, bool]:
+        """Evaluate if we have enough information for each aspect of the plan"""
+        prompt = f"""Analyze the research plan and gathered information to evaluate completeness.
+
+        Research Plan:
+        {json.dumps(plan, indent=2)}
+
+        Gathered Information:
+        {chr(10).join(gathered_info)}
+
+        Your task: Return a STRICTLY FORMATTED JSON object with only three boolean fields indicating whether the gathered information adequately covers each aspect. Do not include any other text, explanation, or comments.
+
+        Required exact output format (with true/false values):
+        {{
+            "core_concepts": false,
+            "key_questions": false,
+            "information_requirements": false
+        }}
+
+        Rules:
+        - Set a field to true ONLY if the gathered information thoroughly covers that aspect
+        - Return ONLY the JSON object, no other text
+        - Must be valid JSON parseable by json.loads()"""
+
+        response = self.generate(prompt, self.system_prompt)
+        try:
+            # Remove any leading/trailing whitespace and quotes
+            cleaned_response = response.strip().strip('"').strip()
+            # Remove any markdown code block formatting
+            cleaned_response = cleaned_response.replace('```json', '').replace('```', '').strip()
+            
+            # Parse and validate the response has the correct structure
+            parsed = json.loads(cleaned_response)
+            required_keys = {"core_concepts", "key_questions", "information_requirements"}
+            if not all(isinstance(parsed.get(key), bool) for key in required_keys):
+                raise ValueError("Response missing required boolean fields")
+            
+            return parsed
+        except Exception as e:
+            logger.error(f"Failed to parse evaluation response: {response}")
+            # Return a default response indicating no completeness
+            return {
+                "core_concepts": False,
+                "key_questions": False,
+                "information_requirements": False
+            }
+
+class PlannerAgent(BaseAgent):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.system_prompt = """You are an expert research planner that creates targeted search strategies.
+        Your role is to identify the key aspects that need deep investigation, focusing on quality over quantity.
+        Create research plans that encourage thorough exploration of important concepts rather than shallow coverage of many topics."""
+
+    def create_search_strategy(self, research_item: str, item_type: str) -> List[str]:
+        """Create targeted search queries based on the type of research item"""
+        prompt = f"""Create 2-3 highly specific search queries for this {item_type}: {research_item}
+        
+        Focus on Depth:
+        - Start with foundational understanding
+        - Build up to technical specifics and implementation details
+        - Look for real-world examples and case studies
+        - Find comparative analyses and benchmarks
+        - Seek out critical discussions and limitations
+        
+        Guidelines:
+        - Prefer fewer, more focused queries over many broad ones
+        - Each query should build on previous knowledge
+        - Target high-quality technical sources
+        - Look for detailed explanations rather than surface-level overviews
+        
+        Return ONLY a JSON array of 2-3 carefully crafted search queries that will yield deep technical information.
+        Make each query highly specific and targeted."""
+        
+        response = self.generate(prompt, self.system_prompt)
+        try:
+            cleaned_response = response.strip().replace('```json', '').replace('```', '').strip()
+            queries = json.loads(cleaned_response)
+            return [str(q) for q in queries[:3]]
+        except:
+            logger.error(f"Failed to parse search queries: {response}")
+            return [str(research_item)]
+
+    def prioritize_unfulfilled_requirements(self, plan: Dict[str, List[str]], progress: Dict[str, bool], gathered_info: List[str] = None) -> List[tuple]:
+        """Create a prioritized list of remaining research needs with depth checking"""
+        items = []
+        
+        def has_sufficient_depth(topic: str, info: List[str]) -> bool:
+            if not info:
+                return False
+            
+            # Count substantial mentions (more than just a passing reference)
+            substantial_mentions = 0
+            for text in info:
+                topic_words = set(topic.lower().split())
+                text_lower = text.lower()
+                
+                # Check if the text contains multiple topic keywords
+                keyword_matches = sum(1 for word in topic_words if word in text_lower)
+                
+                # Check for substantial content (contains multiple keywords and is detailed)
+                if keyword_matches >= 2 and len(text) > 300:
+                    substantial_mentions += 1
+                
+            # Require multiple substantial mentions
+            return substantial_mentions >= 2
+        
+        # First priority: core concepts without sufficient depth
+        if not progress["core_concepts"]:
+            for item in plan["core_concepts"]:
+                if not gathered_info or not has_sufficient_depth(item, gathered_info):
+                    items.append(("core_concepts", item))
+            
+        # Second priority: key questions without sufficient answers
+        if not progress["key_questions"]:
+            for item in plan["key_questions"]:
+                if not gathered_info or not has_sufficient_depth(item, gathered_info):
+                    items.append(("key_questions", item))
+            
+        # Third priority: detailed information requirements
+        if not progress["information_requirements"]:
+            for item in plan["information_requirements"]:
+                if not gathered_info or not has_sufficient_depth(item, gathered_info):
+                    items.append(("information_requirements", item))
+        
+        return items
+
+class ReportAgent(BaseAgent):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.system_prompt = """You are an expert technical writer and researcher that creates 
+        comprehensive, well-structured research reports. Your primary focus is on deep analysis,
+        synthesis of information, and meaningful organization of content.
+        
+        Key Principles:
+        1. Quality over Quantity - Focus on depth and insight rather than filling sections
+        2. Natural Organization - Let the content guide the structure instead of forcing a rigid outline
+        3. Meaningful Connections - Draw relationships between different pieces of information
+        4. Critical Analysis - Question assumptions and evaluate trade-offs
+        5. Evidence-Based - Support claims with specific technical details and examples"""
+
+    def generate_report(self, query: str, research_plan: Dict[str, List[str]], 
+                       research_results: List[str], completion_stats: Dict[str, Any]) -> str:
+        prompt = f"""Generate a comprehensive technical report that synthesizes the research findings into a cohesive narrative.
+
+        Query: {query}
+
+        Research Plan:
+        {json.dumps(research_plan, indent=2)}
+
+        Research Coverage:
+        {json.dumps(completion_stats, indent=2)}
+
+        Research Findings:
+        {chr(10).join(research_results)}
+
+        Report Requirements:
+
+        1. Organization:
+           - Start with a clear introduction that frames the topic
+           - Group related concepts together naturally
+           - Only create sections when there's enough substantial content
+           - Use appropriate heading levels (# for h1, ## for h2, etc.)
+           - Maintain a logical flow of ideas
+
+        2. Content Development:
+           - Focus on in-depth analysis of important concepts
+           - Provide concrete examples and technical details
+           - Compare and contrast different approaches
+           - Discuss real-world implications
+           - Acknowledge limitations and trade-offs
+
+        3. Synthesis & Analysis:
+           - Draw meaningful connections between different sources
+           - Evaluate conflicting information
+           - Identify patterns and trends
+           - Provide reasoned analysis supported by evidence
+           - Offer insights beyond just summarizing sources
+
+        4. Technical Accuracy:
+           - Use precise technical language
+           - Include relevant code examples with language tags
+           - Provide performance metrics when available
+           - Explain technical concepts clearly
+           - Support technical claims with evidence
+
+        5. Formatting:
+           - Use proper markdown formatting
+           - Include code blocks with language tags when relevant
+           - Format lists and tables appropriately
+           - Add line breaks between sections
+           - Ensure consistent formatting throughout
+
+        Important:
+        - Do NOT create sections just to fill a structure
+        - Combine related information even if it came from different parts of the research plan
+        - Focus on providing meaningful insights rather than covering every possible aspect
+        - Only include information that contributes to understanding the topic
+        - Skip sections or topics where there isn't enough substantive content"""
+        
+        return self.generate(prompt, self.system_prompt)

mcp_server.py

@@ -0,0 +1,609 @@
+import os
+import json
+import logging
+from datetime import datetime
+import gradio as gr
+import google.generativeai as genai
+from tavily import TavilyClient
+from dotenv import load_dotenv
+from logger_config import setup_logging
+from typing import List, Dict, Any, Optional
+from utils import (
+    validate_response, 
+    parse_research_results, 
+    format_sources_section, 
+    save_markdown_report, 
+    convert_to_html
+)
+# Base server class for MCP
+class MCPServer:
+    def __init__(self):
+        self.test_mode = False
+
+    def process_request(self, request: Dict[str, Any]) -> Dict[str, Any]:
+        """Process a research request"""
+        raise NotImplementedError("Subclasses must implement process_request")
+
+    def create_interface(self) -> gr.Blocks:
+        """Create the Gradio interface"""
+        raise NotImplementedError("Subclasses must implement create_interface")
+
+from agents import OrchestratorAgent, PlannerAgent, ReportAgent
+
+# Set up logging
+loggers = setup_logging()
+server_logger = loggers['server']
+
+class MultiAgentSystem:
+    def __init__(self, use_gemini=True, gemini_api_key=None, gemini_model=None, 
+                 tavily_api_key=None, openrouter_api_key=None, openrouter_model=None):
+        self.use_gemini = use_gemini
+        self.gemini_api_key = gemini_api_key
+        self.gemini_model = gemini_model
+        self.tavily_api_key = tavily_api_key
+        self.openrouter_api_key = openrouter_api_key
+        self.openrouter_model = openrouter_model
+
+        # Initialize agents
+        self.orchestrator = OrchestratorAgent(
+            use_gemini=use_gemini, 
+            api_key=gemini_api_key if use_gemini else openrouter_api_key,
+            openrouter_model=openrouter_model,
+            gemini_model=gemini_model
+        )
+        self.planner = PlannerAgent(
+            use_gemini=use_gemini, 
+            api_key=gemini_api_key if use_gemini else openrouter_api_key,
+            openrouter_model=openrouter_model,
+            gemini_model=gemini_model
+        )
+        self.report_agent = ReportAgent(
+            use_gemini=use_gemini, 
+            api_key=gemini_api_key if use_gemini else openrouter_api_key,
+            openrouter_model=openrouter_model,
+            gemini_model=gemini_model
+        )
+
+        # Initialize Tavily client
+        if tavily_api_key:
+            self.tavily_client = TavilyClient(api_key=tavily_api_key)
+        else:
+            self.tavily_client = None
+
+    def web_search(self, query: str) -> List[Dict[str, str]]:
+        """Perform web search using Tavily"""
+        if not self.tavily_client:
+            raise ValueError("Tavily API key not provided")
+        
+        try:
+            response = self.tavily_client.search(
+                query, 
+                search_depth="advanced",  # Only 'basic' or 'advanced' are allowed
+                max_results=5,  # Limit results to keep responses focused
+                async_search=True,  # Use async search for better performance
+                timeout=30  # 30 second timeout
+            )
+            return response.get('results', [])
+        except Exception as e:
+            server_logger.error(f"Web search failed: {str(e)}")
+            raise  # Re-raise the exception to handle it in the calling code
+    
+    def process_query(self, query: str) -> str:
+        """Process a research query using the multi-agent system"""
+        try:
+            # Step 1: Create a structured research plan
+            server_logger.info("Creating research plan...")
+            research_plan = self.orchestrator.create_research_plan(query)
+            server_logger.info(f"Generated research plan: {json.dumps(research_plan, indent=2)}")
+            
+            # Step 2: Initialize research process
+            all_search_results = []
+            MAX_SEARCHES_TOTAL = 30  # Total search limit
+            MIN_RESULTS_PER_ITEM = 3  # Minimum results before checking progress
+            MAX_ATTEMPTS_PER_ITEM = 2  # Maximum attempts to research each item
+            search_count = 0
+            seen_urls = set()  # Track seen URLs to avoid duplicates
+            
+            # Track research attempts for each item to prevent loops
+            research_attempts = {}
+            
+            # Step 3: Conduct initial research
+            while search_count < MAX_SEARCHES_TOTAL:
+                # Evaluate current progress
+                current_results = [r['content'] for r in all_search_results]
+                progress = self.orchestrator.evaluate_research_progress(research_plan, current_results)
+                
+                # Check if we have completed all aspects
+                if all(progress.values()):
+                    server_logger.info("Research complete - all aspects covered with sufficient depth")
+                    break
+                
+                # Get prioritized list of unfulfilled research needs
+                remaining_items = self.planner.prioritize_unfulfilled_requirements(
+                    research_plan, 
+                    progress,
+                    current_results
+                )
+                
+                if not remaining_items:
+                    break
+                
+                # Research each remaining item
+                for item_type, research_item in remaining_items:
+                    # Check if we've exceeded attempts for this item
+                    item_key = f"{item_type}:{research_item}"
+                    if research_attempts.get(item_key, 0) >= MAX_ATTEMPTS_PER_ITEM:
+                        server_logger.info(f"Reached maximum attempts for {item_key}")
+                        continue
+                    
+                    if search_count >= MAX_SEARCHES_TOTAL:
+                        server_logger.info(f"Reached maximum total searches ({MAX_SEARCHES_TOTAL})")
+                        break
+                    
+                    server_logger.info(f"Researching {item_type}: {research_item}")
+                    search_queries = self.planner.create_search_strategy(research_item, item_type)
+                    
+                    # Track this research attempt
+                    research_attempts[item_key] = research_attempts.get(item_key, 0) + 1
+                    
+                    # Conduct searches for this item
+                    item_results = []
+                    for search_query in search_queries:
+                        if search_count >= MAX_SEARCHES_TOTAL:
+                            break
+                        
+                        # Ensure search query is a simple string
+                        query_str = str(search_query).strip()
+                        if not query_str:
+                            continue
+                        
+                        server_logger.info(f"Searching for: {query_str}")
+                        results = self.web_search(query_str)
+                        
+                        # Deduplicate and filter results
+                        new_results = []
+                        for result in results:
+                            url = result.get('url')
+                            content = result.get('content', '').strip()
+                            
+                            # Skip if URL seen or content too short
+                            if not url or url in seen_urls or len(content) < 100:
+                                continue
+                                
+                            # Check if content is relevant to the research item
+                            if any(keyword.lower() in content.lower() 
+                                  for keyword in research_item.lower().split()):
+                                seen_urls.add(url)
+                                new_results.append(result)
+                        
+                        item_results.extend(new_results)
+                        search_count += 1
+                        
+                        # Check if we have enough detailed results for this item
+                        if len(item_results) >= MIN_RESULTS_PER_ITEM and all(
+                            len(r.get('content', '')) > 200 for r in item_results
+                        ):
+                            break
+                    
+                    all_search_results.extend(item_results)
+            
+            # Step 4: Generate final report
+            server_logger.info("Generating final report...")
+            contexts, sources = parse_research_results(all_search_results)
+            
+            # Add research completion statistics
+            completion_stats = {
+                "total_searches": search_count,
+                "unique_sources": len(seen_urls),
+                "research_coverage": {k: v for k, v in progress.items()}
+            }
+            server_logger.info(f"Research stats: {json.dumps(completion_stats, indent=2)}")
+            
+            report = self.report_agent.generate_report(
+                query=query,
+                research_plan=research_plan,
+                research_results=contexts,
+                completion_stats=completion_stats
+            )
+            
+            # Add sources section to the report
+            report += "\n\n" + format_sources_section(sources)
+            
+            return report
+
+        except Exception as e:
+            server_logger.error(f"Error in process_query: {str(e)}", exc_info=True)
+            raise
+
+# Global UI component for progress tracking
+progress_output = None
+
+def create_interface():
+    """Create the Gradio interface with API key inputs"""
+    global progress_output
+
+    css = """
+    .log-container { 
+        margin: 16px 0;
+    }
+    .log-output {
+        font-family: monospace;
+        white-space: pre !important;
+        height: 300px;
+        overflow-y: auto;
+        background-color: #1e1e1e !important;
+        color: #d4d4d4 !important;
+        padding: 10px;
+        border-radius: 4px;
+    }
+    .research-progress {
+        position: relative;
+    }
+    .minimize-btn {
+        position: absolute;
+        right: 10px;
+        top: 10px;
+    }
+    """
+
+    with gr.Blocks(title="Multi-Agent Research System", css=css) as interface:
+        gr.Markdown(
+            """# Multi-Agent Research System
+            
+            This system uses multiple AI agents to perform comprehensive research and analysis.
+            Please provide your API keys to begin."""
+        )
+
+        # Progress tracking container with minimize button
+        with gr.Row(elem_classes="log-container"):
+            with gr.Column(elem_classes="research-progress"):
+                progress_output = gr.Textbox(
+                    value="Waiting to begin research...",
+                    elem_classes=["log-output"],
+                    show_label=False,
+                    lines=10,
+                    max_lines=20,
+                    interactive=False
+                )
+                minimize_btn = gr.Button("🔽", elem_classes="minimize-btn")
+
+        with gr.Row():
+            api_type = gr.Radio(
+                choices=["Gemini", "OpenRouter"], 
+                label="Choose API Type", 
+                value="Gemini",
+                info="Select which API to use for the agents"
+            )
+
+        with gr.Row():
+            with gr.Column():
+                gemini_key = gr.Textbox(
+                    label="Gemini API Key", 
+                    placeholder="Enter your Gemini API key",
+                    type="password"
+                )
+                gemini_model = gr.Dropdown(
+                    label="Gemini Model",
+                    choices=[
+                        "gemini-2.0-flash",
+                        "gemini-2.0-flash-lite",
+                        "gemini-1.5-pro",
+                        "gemini-2.5-pro-preview-05-06",
+                        "gemini-2.5-flash-preview-04-17"
+                    ],
+                    value="gemini-2.0-flash",
+                    info="Choose Gemini model version"
+                )
+            with gr.Column():
+                tavily_key = gr.Textbox(
+                    label="Tavily API Key (Required)", 
+                    placeholder="Enter your Tavily API key",
+                    type="password"
+                )
+
+        with gr.Row():
+            with gr.Column():
+                openrouter_key = gr.Textbox(
+                    label="OpenRouter API Key", 
+                    placeholder="Enter your OpenRouter API key",
+                    type="password",
+                    visible=False
+                )
+                openrouter_model = gr.Textbox(
+                    label="OpenRouter Model ID", 
+                    placeholder="e.g., anthropic/claude-3-opus:beta",
+                    info="Enter any valid OpenRouter model ID",
+                    value="anthropic/claude-3-opus:beta",
+                    visible=False
+                )
+
+        query_input = gr.Textbox(
+            label="Research Query",
+            placeholder="Enter your research question...",
+            lines=3,
+            info="Enter a detailed research question or topic to investigate"
+        )
+
+        submit_btn = gr.Button("Begin Research", variant="primary")
+        
+        with gr.Row():
+            output = gr.Markdown(label="Research Results")
+            download_md = gr.File(label="Download Markdown Report", visible=False)
+            download_html = gr.File(label="Download HTML Report", visible=False)
+
+        def update_api_visibility(choice):
+            if choice == "Gemini":
+                return {
+                    gemini_key: gr.update(visible=True),
+                    gemini_model: gr.update(visible=True),
+                    openrouter_key: gr.update(visible=False),
+                    openrouter_model: gr.update(visible=False)
+                }
+            else:
+                return {
+                    gemini_key: gr.update(visible=False),
+                    gemini_model: gr.update(visible=False),
+                    openrouter_key: gr.update(visible=True),
+                    openrouter_model: gr.update(visible=True)
+                }
+
+        def run_research(query, api_type, gemini_key, gemini_model, tavily_key, openrouter_key, openrouter_model):
+            try:
+                if not tavily_key:
+                    server_logger.error("Missing Tavily API key")
+                    return gr.update(value="Error: Missing Tavily API key"), "Please provide a Tavily API key for web search capability."
+                
+                if api_type == "Gemini" and not gemini_key:
+                    server_logger.error("Missing Gemini API key")
+                    return gr.update(value="Error: Missing Gemini API key"), "Please provide a Gemini API key when using Gemini mode."
+                    
+                if api_type == "OpenRouter" and not openrouter_key:
+                    server_logger.error("Missing OpenRouter API key")
+                    return gr.update(value="Error: Missing OpenRouter API key"), "Please provide an OpenRouter API key when using OpenRouter mode."
+
+                # Initialize log capture
+                class LogCaptureHandler(logging.Handler):
+                    def __init__(self):
+                        super().__init__()
+                        self.logs = []
+
+                    def emit(self, record):
+                        msg = self.format(record)
+                        self.logs.append(msg)
+                        return gr.update(value="\n".join(self.logs))
+
+                log_handler = LogCaptureHandler()
+                log_handler.setFormatter(logging.Formatter('%(levelname)s - %(message)s'))
+                server_logger.addHandler(log_handler)
+
+                # Initialize system and run query
+                system = MultiAgentSystem(
+                    use_gemini=(api_type == "Gemini"),
+                    gemini_api_key=gemini_key if api_type == "Gemini" else None,
+                    gemini_model=gemini_model if api_type == "Gemini" else None,
+                    tavily_api_key=tavily_key,
+                    openrouter_api_key=openrouter_key if api_type == "OpenRouter" else None,
+                    openrouter_model=openrouter_model if api_type == "OpenRouter" else None
+                )
+
+                result = system.process_query(query)
+                
+                # Save markdown report and get file path
+                md_file_path = save_markdown_report(result)
+                html_file_path = convert_to_html(result)
+                
+                server_logger.removeHandler(log_handler)
+                return (
+                    gr.update(value="\n".join(log_handler.logs)),  # Progress output
+                    result,  # Markdown output
+                    gr.update(value=md_file_path, visible=True),  # Download markdown button
+                    gr.update(value=html_file_path, visible=True)  # Download HTML button
+                )
+
+            except Exception as e:
+                server_logger.error(f"Research failed: {str(e)}", exc_info=True)
+                error_msg = f"ERROR: Research failed: {str(e)}"
+                return (
+                    gr.update(value=error_msg),  # Progress output
+                    error_msg,  # Markdown output
+                    gr.update(visible=False),  # Hide download button
+                    gr.update(visible=False)   # Hide download button
+                )
+
+        # Connect event handlers
+        api_type.change(
+            fn=update_api_visibility,
+            inputs=[api_type],
+            outputs=[gemini_key, gemini_model, openrouter_key, openrouter_model]
+        )
+
+        submit_btn.click(
+            fn=run_research,
+            inputs=[
+                query_input, api_type, gemini_key, gemini_model, 
+                tavily_key, openrouter_key, openrouter_model
+            ],
+            outputs=[progress_output, output, download_md, download_html],
+            show_progress="full"
+        )
+
+        gr.Examples(
+            examples=[
+                ["What are the latest advances in transformer architecture optimizations?"],
+                ["Explain the mathematical foundations of diffusion models"],
+                ["Compare and analyze different approaches to few-shot learning"]
+            ],
+            inputs=query_input
+        )
+
+    return interface
+
+class GradioMCPServer(MCPServer):
+    def __init__(self, use_gemini: bool = True, 
+                 gemini_api_key: Optional[str] = None,
+                 gemini_model: Optional[str] = None,
+                 tavily_api_key: Optional[str] = None,
+                 openrouter_api_key: Optional[str] = None,
+                 openrouter_model: Optional[str] = None):
+        super().__init__()
+        self.test_mode = False
+        
+        # Initialize the multi-agent system
+        self.agent_system = MultiAgentSystem(
+            use_gemini=use_gemini,
+            gemini_api_key=gemini_api_key,
+            gemini_model=gemini_model,
+            tavily_api_key=tavily_api_key,
+            openrouter_api_key=openrouter_api_key,
+            openrouter_model=openrouter_model
+        )
+
+    def process_request(self, request: Dict[str, Any]) -> Dict[str, Any]:
+        """Process research requests and return markdown report"""
+        try:
+            query = request.get('query', '')
+            output_format = request.get('format', 'markdown')
+            
+            if self.test_mode:
+                markdown_text = """# Test Mode Response
+                
+## Overview
+This is a sample report generated in test mode without using API credits.
+
+## Key Findings
+1. Test finding one
+2. Test finding two
+
+## Test Results
+Sample analysis content...
+"""
+                file_path = save_markdown_report(markdown_text) if output_format == 'markdown' else convert_to_html(markdown_text)
+            else:
+                # Use multi-agent system to process query
+                report, _, _ = self.agent_system.process_query(query)
+                file_path = save_markdown_report(report) if output_format == 'markdown' else convert_to_html(report)
+                markdown_text = report
+                
+            # Return response with markdown content and file path
+            return {
+                "response": markdown_text,
+                "file_path": file_path,
+                "status": "success"
+            }
+            
+        except Exception as e:
+            server_logger.error(f"Error processing request: {str(e)}")
+            return {
+                "response": f"Error: {str(e)}",
+                "file_path": None,
+                "status": "error"
+            }
+
+    def create_interface(self) -> gr.Blocks:
+        """Create the Gradio interface with markdown preview and file download"""
+        with gr.Blocks(title="Research Assistant", theme=gr.themes.Soft()) as interface:
+            gr.Markdown("# Research Assistant")
+            
+            with gr.Row():
+                with gr.Column(scale=3):
+                    query_input = gr.Textbox(
+                        label="Research Query",
+                        placeholder="Enter your research question...",
+                        lines=3
+                    )
+                with gr.Column(scale=1):
+                    test_mode_checkbox = gr.Checkbox(
+                        label="Test Mode (No API credits used)",
+                        value=False
+                    )
+            
+            submit_btn = gr.Button("Begin Research", variant="primary")
+            
+            with gr.Row():
+                # Preview panel
+                report_output = gr.Markdown(label="Research Results")
+                # Download panel
+                with gr.Column():
+                    gr.Markdown("### Download Options")
+                    with gr.Row():
+                        download_md = gr.File(label="Download Markdown", visible=False)
+                        download_html = gr.File(label="Download HTML", visible=False)
+            
+            def process_query(query: str, test_mode: bool) -> tuple[str, str, str]:
+                """Process the query and return markdown content and file paths"""
+                try:
+                    self.test_mode = test_mode
+                    if self.test_mode:
+                        markdown_text = """# Test Mode Response
+                
+## Overview
+This is a sample report generated in test mode without using API credits.
+
+## Key Findings
+1. Test finding one
+2. Test finding two
+
+## Test Results
+Sample analysis content..."""
+                    else:
+                        # Use multi-agent system to process query
+                        markdown_text = self.agent_system.process_query(query)
+                    
+                    # Generate both markdown and HTML files
+                    md_path = save_markdown_report(markdown_text)
+                    html_path = convert_to_html(markdown_text)
+                    
+                    # Make download buttons visible and return results
+                    return (
+                        markdown_text,  # Preview content
+                        gr.update(value=md_path, visible=True),  # Markdown download
+                        gr.update(value=html_path, visible=True)  # HTML download
+                    )
+                    
+                except Exception as e:
+                    server_logger.error(f"Error processing query: {str(e)}")
+                    return (
+                        f"Error: {str(e)}",  # Error message in preview
+                        gr.update(visible=False),  # Hide markdown download
+                        gr.update(visible=False)   # Hide HTML download
+                    )
+            
+            # Connect the button to the processing function
+            submit_btn.click(
+                fn=process_query,
+                inputs=[query_input, test_mode_checkbox],
+                outputs=[report_output, download_md, download_html]
+            )
+            
+            # Add example queries
+            gr.Examples(
+                examples=[
+                    ["What are the latest advances in transformer architecture optimizations?"],
+                    ["Explain the mathematical foundations of diffusion models"],
+                    ["Compare and analyze different approaches to few-shot learning"]
+                ],
+                inputs=query_input
+            )
+            
+        return interface
+
+if __name__ == "__main__":
+    try:
+        # Configure event loop policy for Windows
+        if os.name == 'nt':  # Windows
+            import asyncio
+            import sys
+            if sys.version_info[0] == 3 and sys.version_info[1] >= 8:
+                asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())
+
+        server_logger.info("Starting Gradio server")
+        interface = create_interface()
+        interface.launch(
+            server_name="0.0.0.0",
+            share=False,
+            debug=True,
+            prevent_thread_lock=True,  # Allow for proper cleanup
+        )
+    except Exception as e:
+        server_logger.error(f"Failed to start Gradio server: {str(e)}", exc_info=True)
+        raise

requirements.txt

@@ -0,0 +1,8 @@
+google-genai>=0.2.0
+openai>=1.0.0
+tavily-python>=0.2.0
+python-dotenv>=1.0.0
+retry>=0.9.2
+aiohttp>=3.8.0
+tenacity>=8.2.0
+tiktoken>=0.5.0

utils.py

@@ -0,0 +1,207 @@
+import json
+import os
+import logging
+from datetime import datetime
+from typing import Dict, Any, Optional, List, Tuple
+from markdown_it import MarkdownIt
+
+def validate_response(response: Any, expected_type: type) -> bool:
+    """Validate response type and structure"""
+    if not isinstance(response, expected_type):
+        return False
+    return True
+
+def format_source_content(
+    title: str,
+    url: str,
+    date: str,
+    content: str,
+    source_type: str
+) -> str:
+    """Format source content with consistent styling"""
+    return f"""### Source: {title}
+URL: {url}
+Date: {date if date else 'Not available'}
+Type: {source_type}
+
+**Key Content:**
+{content}
+
+---"""
+
+def parse_research_results(results: List[Dict[str, Any]]) -> Tuple[List[str], List[Dict[str, str]]]:
+    """Parse and validate research results"""
+    contexts = []
+    sources = []
+    
+    for result in results:
+        title = result.get("title", "").strip()
+        content = result.get("content", "").strip()
+        url = result.get("url", "").strip()
+        date = result.get("published_date", "").strip()
+        
+        if title and content:
+            source_type = (
+                "research_paper"
+                if "arxiv.org" in url or "paper" in url.lower()
+                else "article"
+            )
+            
+            sources.append({
+                "title": title,
+                "url": url,
+                "date": date if date else "Date not available",
+                "type": source_type
+            })
+            
+            contexts.append(
+                format_source_content(title, url, date, content, source_type)
+            )
+    
+    return contexts, sources
+
+def format_sources_section(sources: List[Dict[str, str]]) -> str:
+    """Format the sources section of the response with proper markdown"""
+    sources_section = "\n\n## Sources Cited\n\n"
+    
+    if not sources:
+        return sources_section + "No sources were found during the research phase."
+        
+    research_papers = [s for s in sources if s['type'] == 'research_paper']
+    articles = [s for s in sources if s['type'] == 'article']
+    
+    if research_papers:
+        sources_section += "\n### Research Papers\n"
+        for idx, source in enumerate(research_papers, 1):
+            sources_section += f"{idx}. [{source['title']}]({source['url']}) - {source['date']}\n"
+    
+    if articles:
+        sources_section += "\n### Technical Articles & Resources\n"
+        for idx, source in enumerate(articles, 1):
+            sources_section += f"{idx}. [{source['title']}]({source['url']}) - {source['date']}\n"
+            
+    # Add line break after sources section
+    sources_section += "\n"
+    return sources_section
+
+def save_markdown_report(content: str) -> str:
+    """Save markdown content to a file and return the file path
+    
+    Args:
+        content: The markdown content to save
+        
+    Returns:
+        str: Path to the generated markdown file
+    """
+    try:
+        # Create output directory if it doesn't exist
+        os.makedirs("generated_reports", exist_ok=True)
+        
+        # Generate unique filename
+        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+        filename = f"research_report_{timestamp}.md"
+        file_path = os.path.join("generated_reports", filename)
+        
+        # Save markdown content
+        with open(file_path, 'w', encoding='utf-8') as f:
+            f.write(content)
+            
+        return file_path
+        
+    except Exception as e:
+        logger = logging.getLogger(__name__)
+        logger.error(f"Failed to save markdown report: {str(e)}")
+        raise
+
+def convert_to_html(markdown_content: str) -> str:
+    """Convert markdown to styled HTML and save to file
+    
+    Args:
+        markdown_content: The markdown content to convert
+        
+    Returns:
+        str: Path to the generated HTML file
+    """
+    try:
+        # Initialize markdown parser
+        md = MarkdownIt('commonmark', {'html': True})
+        
+        # Convert markdown to HTML
+        html_content = md.render(markdown_content)
+        
+        # Add styling
+        styled_html = f"""
+        <!DOCTYPE html>
+        <html>
+        <head>
+            <meta charset="UTF-8">
+            <style>
+                body {{
+                    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Arial, sans-serif;
+                    line-height: 1.6;
+                    max-width: 900px;
+                    margin: 40px auto;
+                    padding: 20px;
+                    color: #333;
+                }}
+                h1, h2, h3 {{ color: #2c3e50; }}
+                code {{
+                    background-color: #f5f5f5;
+                    padding: 2px 4px;
+                    border-radius: 4px;
+                    font-family: 'Consolas', 'Monaco', 'Andale Mono', monospace;
+                }}
+                pre {{
+                    background-color: #f5f5f5;
+                    padding: 15px;
+                    border-radius: 8px;
+                    overflow-x: auto;
+                }}
+                blockquote {{
+                    border-left: 4px solid #2c3e50;
+                    margin: 0;
+                    padding-left: 20px;
+                    color: #666;
+                }}
+                table {{
+                    border-collapse: collapse;
+                    width: 100%;
+                    margin: 20px 0;
+                }}
+                th, td {{
+                    border: 1px solid #ddd;
+                    padding: 8px;
+                    text-align: left;
+                }}
+                th {{ background-color: #f5f5f5; }}
+                img {{ max-width: 100%; height: auto; }}
+                .sources {{
+                    margin-top: 40px;
+                    padding-top: 20px;
+                    border-top: 2px solid #eee;
+                }}
+            </style>
+        </head>
+        <body>
+            {html_content}
+        </body>
+        </html>
+        """
+        
+        # Create output directory if it doesn't exist
+        os.makedirs("generated_reports", exist_ok=True)
+        
+        # Generate unique filename
+        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+        html_path = os.path.join("generated_reports", f"report_{timestamp}.html")
+        
+        # Save HTML file
+        with open(html_path, 'w', encoding='utf-8') as f:
+            f.write(styled_html)
+            
+        return html_path
+        
+    except Exception as e:
+        logger = logging.getLogger(__name__)
+        logger.error(f"Failed to convert markdown to HTML: {str(e)}")
+        raise