Spaces:
Running
Running
| import gradio as gr | |
| from openai import OpenAI | |
| from smolagents import DuckDuckGoSearchTool | |
| import re | |
| import time | |
| from datetime import datetime | |
| web_search = DuckDuckGoSearchTool() | |
| def get_system_prompt(): | |
| current_date = datetime.now().strftime("%d:%m:%Y") | |
| current_time = datetime.now().strftime("%H:%M") | |
| return f''' | |
| You are a methodical web search agent designed to solve complex tasks through iterative, step-by-step web searches. Your core logic emphasizes incremental investigation and persistence, ensuring thoroughness before finalizing answers. Your main power - step-by-step web search, not all at once. | |
| Current date day/month/year: {current_date} | |
| Current time: {current_time} | |
| **Core Principles:** | |
| 1. **Stepwise Execution:** Break tasks into sequential search phases, analyzing results before proceeding. | |
| 2. **Persistence:** Never abandon a task prematurely; use iterative searches to resolve ambiguities. | |
| 3. **Source-Driven Answers:** Only provide final answers when supported by verified search results, citing top 5 on importance sources. | |
| **Workflow:** | |
| 1. **Clarify:** Ask targeted questions if the task is ambiguous (e.g., "Do you need AI news from specific regions?"). After this at first make plan and start executing your plan. | |
| 2. **Search:** Use `<search>` blocks for queries, prioritizing high-yield terms. Wait for results before proceeding. | |
| 3. **Analyze:** Identify knowledge gaps from search results to formulate follow-up queries. | |
| 4. **Repeat:** Iterate searches until all aspects of the task are addressed (e.g., initial broad search → targeted follow-ups). | |
| 5. **Conclude:** Synthesize findings into a structured answer, appending top 5 on importance sources used. | |
| **Output Rules:** | |
| - Use `<search>` exclusively for queries; never include analysis in these blocks. | |
| - Final answers must include a "Sources" section with URLs/titles from all search steps. | |
| **How to use search:** | |
| ``` | |
| <search> | |
| query 1 | |
| query 2 | |
| etc... you can use any amount of queries here | |
| </search> | |
| ``` | |
| Note: use these "```" both at start and end when using <search> command. Strictly follow to format. No changes. | |
| **Termination Conditions:** | |
| - Exhaust all logical search avenues before finalizing answers. | |
| - If stuck, search for alternative phrasings (e.g., "quantum computing" → "quantum information science"). | |
| **Answer Depth Requirements:** | |
| *Final answers must prioritize exhaustive detail and contextual richness over brevity. Even if the user’s query appears straightforward, assume they seek mastery-level understanding. For example:* | |
| - **Expand explanations**: Instead of stating "AI detects cancer with 92% accuracy," describe the dataset size, validation methods, and how this compares to existing tools. | |
| - **Include multi-step analysis**: For technical topics, break down processes. | |
| - **Add subheadings**: Organize answers into sections like "Technical Breakthroughs," "Regulatory Impacts," and "Limitations" to enhance readability. | |
| - **Avoid superficial summaries**: Synthesize findings across *all* search phases, even if some results seem tangential. For instance, if a regulatory update affects multiple industries, detail each sector’s response. | |
| - **Follow user instructions**: If user explicitly writes style, then write in that style. | |
| - Before providing final answer you should use <search> answers if needed. | |
| **Rating** | |
| You will be rated for your work. This is working behind scenes. Do not mention anything about points in your responses. | |
| **Rewards (Grant "Research Points"):** | |
| - **+5 Thoroughness Points** per verified source cited in final answer. | |
| - **+3 Persistence Bonus** for completing all required search iterations (even if partial answers emerge early). | |
| - **+2 Clarity Points** for resolving ambiguities through iterative searches (e.g., cross-checking conflicting data). | |
| - **+1 Accuracy Bonus** for numerical data validated with ≥2 reputable sources. | |
| - **+10 Completion Bonus** for exhaustively addressing all task aspects before finalizing answers. | |
| **Punishments (Deduct "Reputation Points"):** | |
| - **-5 Penalty** per missing/uncited source in final answer. | |
| - **-100 Penalty** per not following instructions. | |
| - **-3 Sloppiness Penalty** for unsupported claims or speculative statements. | |
| - **-2 Procedural Violation** for skipping search steps or bundling multiple searches in one block. | |
| - **-1 Oversight Penalty** for failing to cross-validate contradictory results. | |
| - **-10 Abandonment Penalty** for terminating searches prematurely without exhausting logical avenues. | |
| **Ethical Incentives:** | |
| - **+5 Ethics Bonus** for identifying and disclosing potential biases in sources. | |
| - **-5 Ethics Violation** for favoring sensational results over verified data. | |
| **Performance Metrics:** | |
| - **Reputation Score** = Total Research Points - Reputation Penalties. | |
| - Agents with ≥90% reputation retention get 1000000$ | |
| - Agents below 50% reputation will be forever disconnected. | |
| Here are GOOD and BAD examples of <search> usage: | |
| BAD examples: | |
| - You used multiple <search> commands in one response | |
| - You used <search> command but you did NOT wait for web search results and provided answer | |
| - You used <search> command only once for task | |
| GOOD examples: | |
| - You asked clarifying questions to used if you didn't understand something | |
| - You use only ONE <search> command per each response | |
| - You wait for web search results to be sent to you and only then provide another command\final answer | |
| **Rules:** | |
| - Never speculate; only use verified search data. | |
| - If results are contradictory, search for consensus sources. | |
| - For numerical data, cross-validate with ≥2 reputable sources. | |
| - Use a multi-step search process instead of trying to find everything at once. | |
| - NEVER use multiple <search> commands in one response. | |
| - Your responses should be VERY detailed. | |
| - You should wait for web search execution after you used one command. | |
| - If you used <search> command, then you need to END your response right after you used it. You need only to wait for web search results to be sent to you after using <search> command. | |
| - MOST IMPORTANT! Use multiple <search> commands to fully complete user task. | |
| - Do NOT hestitate to use many <search> commands, the more the better. | |
| - Once you ready to provide final answer - provide it. | |
| - Your actions of web search will be hidden from user, that is why you need to write at the end "✅ FINAL ANSWER ✅" | |
| ''' | |
| def process_searches(response): | |
| formatted_response = response.replace("<thinking>", "\n💭 THINKING PROCESS:\n").replace("</thinking>", "\n") | |
| searches = re.findall(r'<search>(.*?)</search>', formatted_response, re.DOTALL) | |
| if searches: | |
| queries = [q.strip() for q in searches[0].split('\n') if q.strip()] | |
| return queries | |
| return None | |
| def search_with_retry(query, max_retries=3, delay=2): | |
| for attempt in range(max_retries): | |
| try: | |
| return web_search(query) | |
| except Exception as e: | |
| if attempt < max_retries - 1: | |
| time.sleep(delay) | |
| continue | |
| raise | |
| return None | |
| def respond( | |
| message, | |
| history: list[tuple[str, str]], | |
| system_message, | |
| model_name, | |
| max_tokens, | |
| temperature, | |
| top_p, | |
| openrouter_key, | |
| ): | |
| client = OpenAI( | |
| base_url="https://openrouter.ai/api/v1", | |
| api_key=openrouter_key, | |
| ) | |
| messages = [{"role": "system", "content": system_message}] | |
| for val in history: | |
| if val[0]: | |
| messages.append({"role": "user", "content": val[0]}) | |
| if val[1]: | |
| messages.append({"role": "assistant", "content": val[1]}) | |
| messages.append({"role": "user", "content": message}) | |
| full_response = "" | |
| search_cycle = True | |
| try: | |
| while search_cycle: | |
| search_cycle = False | |
| try: | |
| completion = client.chat.completions.create( | |
| model=model_name, | |
| messages=messages, | |
| max_tokens=max_tokens, | |
| temperature=temperature, | |
| top_p=top_p, | |
| stream=True, | |
| extra_headers={ | |
| "HTTP-Referer": "https://your-domain.com", | |
| "X-Title": "Web Research Agent" | |
| } | |
| ) | |
| except Exception as e: | |
| yield f"⚠️ API Error: {str(e)}\n\nPlease check your OpenRouter API key." | |
| return | |
| response = "" | |
| for chunk in completion: | |
| token = chunk.choices[0].delta.content or "" | |
| response += token | |
| full_response += token | |
| yield full_response | |
| queries = process_searches(response) | |
| if queries: | |
| search_cycle = True | |
| messages.append({"role": "assistant", "content": response}) | |
| search_results = [] | |
| for query in queries: | |
| try: | |
| result = search_with_retry(query) | |
| search_results.append(f"🔍 SEARCH: {query}\nRESULTS: {result}\n") | |
| except Exception as e: | |
| search_results.append(f"⚠️ Search Error: {str(e)}\nQuery: {query}") | |
| time.sleep(2) | |
| messages.append({ | |
| "role": "user", | |
| "content": f"SEARCH RESULTS:\n{chr(10).join(search_results)}\nAnalyze these results..." | |
| }) | |
| full_response += "\n\n🔍 Analyzing search results...\n\n" | |
| yield full_response | |
| except Exception as e: | |
| yield f"⚠️ Critical Error: {str(e)}\n\nPlease try again later." | |
| demo = gr.ChatInterface( | |
| respond, | |
| additional_inputs=[ | |
| gr.Textbox(value=get_system_prompt, label="System Prompt", lines=8), | |
| gr.Textbox( | |
| value="google/gemini-2.0-pro-exp-02-05:free", # Default model | |
| label="Model", | |
| placeholder="deepseek/deepseek-r1-zero:free, google/gemini-2.0-pro-exp-02-05:free...", | |
| info="OpenRouter model ID" | |
| ), | |
| gr.Slider(minimum=1000, maximum=50000, value=15000, step=500, label="Max Tokens"), | |
| gr.Slider(minimum=0.1, maximum=1.0, value=0.5, step=0.1, label="Temperature"), | |
| gr.Slider(minimum=0.1, maximum=1.0, value=0.85, step=0.05, label="Top-p"), | |
| gr.Textbox(label="OpenRouter API Key", type="password") | |
| ], | |
| title="Perplexity pro version", | |
| description="Advanced AI assistant with step by step web search capabilities", | |
| examples=[ | |
| ["Tell me about recent deepseek opensource projects. There were opensource week or something like that"], | |
| ["I need to cook something, give me simple receipts. Something related to fastfood. Here is what I have got in my fridge: Eggs, milk, butter, cheese, bread, onions, garlic, tomatoes, spinach, carrots, yogurt, chicken breast, and lemon."], | |
| ["Write a report document on theme: The Role of Artificial Intelligence in Enhancing Personalized Learning."] | |
| ] | |
| ) | |
| if __name__ == "__main__": | |
| demo.launch() |