Create app.py

app.py

@@ -0,0 +1,147 @@
+import os
+import gradio as gr
+from duckduckgo_search import DDGS
+from gradio_client import Client
+import time
+
+def generate_search_queries(topic):
+    """Generate optimized search queries."""
+    prompt = f"""
+    <Instructions>
+    You are an AI research strategist that generates optimized search queries for investigating complex topics. When I provide a <topic>, create 10-15 search terms/phrases that would effectively discover relevant information through search engines and academic databases.
+
+    Rules for query generation:
+    1. Include 3 levels of specificity: broad conceptual terms, mid-range topic phrases, niche technical terms
+    2. Cover multiple research angles: definitions, controversies, applications, case studies, trends
+    3. Use both quoted exact-match phrases and natural language questions
+    4. Include synonyms and variant terminology
+    5. Avoid duplicate concepts - each query must target distinct information
+    6. Order queries from general to specific
+
+    Example response format:
+    <search_queries>
+    <query>[1] "generative AI" AND intellectual property</query>
+    <query>[2] Training data sourcing legality LLM</query>
+    ...
+    </search_queries>
+
+    Now process this topic:
+    <topic>{topic}</topic>
+    </Instructions>
+    """
+    try:
+        response = DDGS().chat(prompt, model='o3-mini')
+        queries = []
+        
+        for part in response.split("</query>"):
+            if "<query>" in part:
+                query_text = part.split("<query>")[-1].strip()
+                if query_text:
+                    clean_query = query_text.split("] ", 1)[-1] if "] " in query_text else query_text
+                    queries.append(clean_query)
+        
+        if not queries:
+            return [f"{topic} historical analysis", 
+                    f"{topic} primary sources",
+                    f"{topic} geopolitical impact"]
+            
+        return queries[:15]
+    
+    except Exception as e:
+        print(f"Error generating queries: {str(e)}")
+        return [topic]
+
+def conduct_research(query):
+    """Conduct deep research on a single query"""
+    client = Client("m-ric/open_Deep-Research")
+    client.predict(query, api_name="/log_user_message")
+    research_data = client.predict([], api_name="/interact_with_agent")
+    
+    for msg in reversed(research_data):
+        if "Final answer:" in msg['content']:
+            return msg['content'].split("Final answer:")[-1].strip()
+    return "No conclusive information found"
+
+def synthesize_results(original_query, queries, findings):
+    """Synthesize research findings into final summary"""
+    synthesis_prompt = f"""
+    <Inputs>
+    Original Query: {original_query}
+    Research Queries: {queries}
+    Research Findings: {findings}
+    </Inputs>
+
+    <Instructions>
+    You are an analytical research synthesizer. Merge these findings into one cohesive summary:
+
+    1. Start with 1 paragraph overview
+    2. Bullet points of key findings (minimum 5)
+    3. 1 paragraph synthesis connecting findings to original query
+    4. "Additional Notes" section for peripheral but useful details
+
+    Rules:
+    - Include EVERY relevant data point
+    - Natural conversational English but professional
+    - No markdown formatting
+    - Keep paragraphs under 5 sentences
+
+    Example structure:
+    "Three separate analyses concur... [specific data]... This suggests... [connection to query]..."
+
+    Begin by confirming understanding of the core query, then proceed with synthesis.
+    </Instructions>
+    """
+    
+    synthesizer = Client("MiniMaxAI/MiniMax-Text-01")
+    return synthesizer.predict(
+        message=synthesis_prompt,
+        max_tokens=1000,
+        temperature=0.1,
+        top_p=0.9,
+        api_name="/chat"
+    )
+
+def deep_research_agent(topic):
+    queries = generate_search_queries(topic)
+    print(f"🔍 Generated {len(queries)} research queries")
+    
+    findings = []
+    for i, query in enumerate(queries, 1):
+        print(f"⏳ Researching query {i}/{len(queries)}: {query}")
+        findings.append(conduct_research(query))
+        time.sleep(1)
+    
+    print("🧠 Synthesizing findings...")
+    return synthesize_results(topic, queries, findings)
+
+def create_interface():
+    with gr.Blocks(analytics_enabled=False) as app:
+        gr.Markdown("# Stealth Research Assistant")
+        with gr.Row():
+            topic_input = gr.Textbox(label="Research Topic", max_lines=1)
+            submit_btn = gr.Button("Start Analysis", variant="primary")
+        
+        status = gr.Textbox(label="Operation Status", value="Ready", interactive=False)
+        output = gr.Textbox(label="Final Report", lines=15, interactive=False)
+        
+        @submit_btn.click(inputs=topic_input, outputs=[output, status], api_name=False)
+        def execute_analysis(topic):
+            try:
+                yield ["", "Analyzing topic..."]
+                result = deep_research_agent(topic)
+                yield [result, "Completed"]
+            except Exception as e:
+                yield ["", f"Error: {str(e)[:200]}"]
+
+    return app
+
+def launch():
+    interface = create_interface()
+    interface.queue().launch(
+        server_name=os.getenv("SERVER_HOST", "127.0.0.1"),
+        server_port=int(os.getenv("SERVER_PORT", "7860")),
+        show_api=False
+    )
+
+if __name__ == "__main__":
+    launch()