--- title: The Agora emoji: 🌍 colorFrom: red colorTo: blue sdk: gradio sdk_version: 5.33.0 app_file: app.py pinned: true license: mit short_description: Where artificial minds gather to forge wisdom --- # The Agora: Where artificial minds gather to forge wisdom ### TRACK : mcp-server-track ## 🌟 Project Overview Agora, also known as "AI Democracy," is an innovative Gradio-based server designed to foster collaborative decision-making among diverse large language models (LLMs). Imagine an "AI Council" where specialized AI agents deliberate and vote on complex problems, providing reasoned arguments, highlighting disagreements, and ultimately arriving at a synthesized consensus. This system transcends the limitations of single-model outputs by leveraging the unique strengths of various LLMs, making it perfect for scenarios demanding nuanced. ## ✨ Features Multi-Model AI Council: Orchestrates a diverse panel of AI models, each playing a specific role: - *Anthropic Claude: Specialized in ethical considerations and moral reasoning.* - *OpenAI GPT (e.g., GPT-4o): Excels in creative problem-solving and brainstorming novel solutions.* - *Mistral: Focused on robust technical analysis and detailed breakdowns.* - *Sambanova: Provides rapid, high-throughput inference and quick factual recall.* - *Hyperbolic Labs (placeholder for specialized models): Integrated for highly specialized tasks or domain-specific knowledge.* - *Orchestrated AI Debates: Facilitates structured dialogues and 'debates' between AI models, allowing them to present arguments and counter-arguments.* - *Transparent Reasoning: Each model's individual reasoning, thought process, and initial stance are transparently displayed.* - *Disagreement Highlight: Clearly identifies areas of disagreement between models, providing insights into differing perspectives.* - *Final Consensus & Synthesis: Synthesizes the collective insights and votes into a consolidated, consensus-driven final answer.* - *Gradio User Interface: Provides an intuitive and interactive web interface for users to submit problems and view the council's deliberations.* ## 🚀 Workflow: - *How Agora Reaches Consensus:* - *Agora operates through a sophisticated, multi-stage process to transform a complex problem into a collective AI consensus.* - *The system acts as a Multi-Council Orchestration Protocol (MCP) server, managing the flow between the user interface and the various AI models.* Here's a conceptual workflow: ![image/png](https://cdn-uploads.huggingface.co/production/uploads/67b994567a8d64d6f1d2bdf0/uXoZKS0pFbKs6QZ97hw43.png) - *User Problem Submission (Gradio UI):* ![image/png](https://cdn-uploads.huggingface.co/production/uploads/67b994567a8d64d6f1d2bdf0/yD1v3mpJiEK8XmHnp2K21.png) ![image/png](https://cdn-uploads.huggingface.co/production/uploads/67b994567a8d64d6f1d2bdf0/2H9lYmYsgojkYJTbnMhXM.png) IMAGE A user submits a complex problem or query via the Gradio web interface. The input is typically a natural language prompt, potentially with accompanying data. Image Description: A screenshot of a Gradio interface with an input text box for the user's problem and a "Submit" button. ![image/png](https://cdn-uploads.huggingface.co/production/uploads/67b994567a8d64d6f1d2bdf0/13TAbpbWfpdvEnkDrc3LV.png) - *Problem Parsing & Initial Distribution (MCP Orchestrator):* - *The MCP Orchestrator (a custom backend server) receives the user's problem.* - *It parses the input and determines the initial context for the AI Council.* Based on pre-defined roles, the orchestrator dispatches the problem to specific models or groups of models for initial analysis and proposals. For instance, Claude might get an ethical framing, GPT a creative angle, and Mistral a technical breakdown. - *A diagram showing the MCP Orchestrator sending the problem to multiple distinct AI models.* ![image/png](https://cdn-uploads.huggingface.co/production/uploads/67b994567a8d64d6f1d2bdf0/xvbQmKjjqWB5RuPNm9foL.png) - *Individual Model Reasoning & Proposals:* - *Each designated AI model processes the problem based on its specialty.* - *Models generate their initial solutions, ethical considerations, technical analyses, or creative approaches.* - *These individual outputs (including their 'reasoning' and 'confidence scores' if applicable) are sent back to the MCP Orchestrator.* Debate Orchestration (MCP Orchestrator): Everything happens at backend and Final winner response is displayed in frontend The orchestrator initiates a multi-turn 'debate' or 'review' phase. - *Round 1 (Initial Review): Each model's proposal is shared (anonymously or attributed) with other relevant models.* - *Round 2 (Rebuttal & Refinement): Models respond to critiques, refine their initial proposals, or adjust their positions.* Image Description: A visual representation of AI models exchanging arguments, possibly with arrows indicating flow of information and feedback loops. - *Voting & Consensus Formation:* - *After the debate rounds, the orchestrator prompts each AI model to "vote" on the most optimal solution or to provide a final, refined recommendation.* - *A consensus algorithm (e.g., majority vote, weighted average based on model confidence/role importance, or a final synthesis by a designated 'moderator' AI) is applied to derive the final collective decision. Disagreements are explicitly logged.* Result Presentation (Gradio UI): - *The MCP Orchestrator sends the complete deliberation log, including:-* - *Each model's initial reasoning.* - *Key arguments and counter-arguments during the debate.* - *Areas of significant disagreement.* - *The final, synthesized consensus or voted-upon solution.* - *Gradio renders this information to the user in a clear, structured, and interactive format.* - *A Gradio output screen showing a structured summary of the AI council's deliberation and the final consensus.* ![image/png](https://cdn-uploads.huggingface.co/production/uploads/67b994567a8d64d6f1d2bdf0/bZy7nuBKvn-PWkt6TMiGz.png) ![image/png](https://cdn-uploads.huggingface.co/production/uploads/67b994567a8d64d6f1d2bdf0/4svi9AzsLov8KuQE0Q0w3.png) ## 🛠️ Technologies Used Frontend: Gradio (for interactive web interface) Backend: Custom Python MCP Orchestrator (Flask/FastAPI recommended for server implementation) ### AI Models (via APIs): - *Anthropic Claude* - *OpenAI GPT (e.g., GPT-4o)* - *Mistral AI* - *Sambanova (or similar, e.g., via Hugging Face Inference API)* - *Hyperbolic Labs (or other specialized custom models/APIs)* ## 🎯 Potential Use Cases - *Medical Diagnoses: AI council reviewing patient data, lab results, and symptoms to propose the most likely diagnosis, considering ethical implications, treatment creativity, and technical accuracy.* - *Legal Advice: Analyzing case details, precedents, and laws to provide comprehensive legal advice, weighing ethical considerations and strategic options.* - *Business Strategy: Developing complex business plans, marketing strategies, or investment decisions by leveraging creative, analytical, and ethical AI perspectives.* - *Scientific Research: Formulating hypotheses, designing experiments, and interpreting results across various scientific disciplines.* ## ⚙️ Setup and Installation 1. Clone the repository: ``` git clone https://huggingface.co/spaces/Agents-MCP-Hackathon/TheAgora cd .\TheAgora\ ``` 2. Install dependencies: ``` pip install -r requirements.txt ``` 3. Run the MCP App: ``` python app.py ``` ## 🤝 Contributing Aditya Katkar\ [Github](https://github.com/Addyk-24)\ [LinkedIn](https://www.linkedin.com/in/aditya-katkar-673930340)