Agora: Multi-Agent Debate and Consensus Engine, Enabling AI to Make Democratic Decisions Like a Parliament

Agora is an innovative multi-agent debate and consensus engine. It conducts structured debates via three AI agents with distinct roles, detects thinking stagnation, reaches a weighted consensus, and generates executable documents (such as policies and technical specifications). Running on the Ollama local model, it is lightweight and portable, suitable for real-world decision-making scenarios, addressing the limitations of single AI models and simulating the human democratic decision-making process.

Current large language models have four major limitations:

• Knowledge Blind Spots: Boundaries of training data lead to insufficient cognition in specific domains
• Fixed Mindset: Prone to falling into fixed reasoning patterns
• Bias Solidification: Implicit biases in training data are amplified
• Lack of Verification: No external feedback mechanism

Humans solve problems through debate and negotiation. Agora introduces this mechanism into AI systems, allowing multi-agent collaboration to reach consensus.

Agora designs three role-based agents:

Proposer

Initiates debates, proposes initial plans, explains the basis and effects, and responds to questions.

Skeptic

Plays the devil's advocate, identifies loopholes and risks in the proposal, and puts forward alternative plans.

Arbitrator

Objectively evaluates arguments, identifies consensus and differences, and makes timely rulings or promotes discussions.

Role separation covers multiple dimensions of the problem and simulates real-world decision-making processes.

Intelligent Stagnation Detection

Identifies deadlock scenarios: repeated arguments, emotional escalation, topic drift, consensus formation, and triggers strategies such as mandatory ruling or introducing new information.

Weighted Consensus Algorithm

Dynamic adjustment of weights for different roles:

• Arbitrator has the highest judgment weight
• Skeptic has high weight in risk assessment
• Proposer has reference value for feasibility Roles with strong logic have their weights increased.

Agora generates four types of documents:

Policy Document

Includes target scope, implementation measures, responsibility division, and risk response plan.

Technical Specification

Includes architecture diagrams, interface definitions, performance indicators, and test plans.

Configuration File

Such as Docker Compose, Kubernetes manifests, and other directly deployable configurations.

Execution Command

Action instructions, priorities, and resource allocation for emergency scenarios.

Ollama brings four major advantages:

• Local Operation: Sensitive data is not uploaded to the cloud, ensuring security
• Flexible Models: Supports Llama/Mistral/Qwen, etc., choose as needed
• Efficient Resources: Low cost, no network restrictions, available offline
• Easy Deployment: Runs on a single machine, no need for distributed architecture.

Applicable scenarios include:

Enterprise Governance

Product roadmap prioritization, technology selection, organizational adjustment, risk management.

Software Development

Architecture review, code refactoring, technical debt handling, open-source license selection.

Personal Decision-Making

Career planning, investment analysis, study plans, major life decisions.

Public Policy

Effect preview, interest game simulation, crisis response plan, urban planning evaluation.

Agora realizes a paradigm shift: from single-model answers to multi-agent collaborative decision-making. Its significance lies in:

1. Improving decision quality and reducing blind spots
2. Enhancing interpretability; the debate process itself is the reason
3. Establishing a trust mechanism with higher transparency
4. Promoting human-machine collaboration; humans can participate in review

In the future, it will play a role in more fields, making AI decision-making more intelligent and humanized.