AI-Decision-Council

🧠 AI Decision Council

Link to the repo

📚 Table of Contents

Overview

AI Decision Council is a lightweight ASP.NET Core web application that enables structured, multi-agent AI deliberation, producing higher-quality, more defensible outcomes by coordinating multiple AI reviewers under a single decision-making authority. In short: it brings governance, review, and synthesis to AI-generated decisions.

Philosophy

Good decisions don’t come from a single voice. They come from structured disagreement, clear thresholds, and deliberate synthesis. The goal of this project is to make that process explicit and programmable.

Why This Exists

Most AI systems optimize for generation. Real decisions require challenge, critique, and synthesis.

AI Decision Council addresses this gap by:

This is particularly useful for:

Key Features

Quick Start (5 Minutes)

Prerequisites

  1. .NET 8 SDK - Download from https://dotnet.microsoft.com/en-us/download/dotnet/8.0. Run the following command to verify your installed version. 
       dotnet --version
  2. AI Models hosted on a service like Azure AI Foundry, or API Keys for services like OpenAI, xAI, Google (Coming soon), Anthropic (Coming soon) etc.
  3. Optional: Visual Studio, VS Code, or another IDE

Clone the Repository

  git clone https://github.com/shriramnat/AI-Decision-Council.git
cd AI-Decision-Council

Configure authentication for the application

Refer to the wiki for details of setting up Environment variables for authentication with Microsoft or Google accounts.

Note: Optionally, you can disable authentication for development environments if you plan on running the app locally for development purposes. You do this by setting the Authentication.Enabled property to false in the appsettings.json file. Refer to the wiki for more details.

Build the Application

  dotnet restore
  dotnet build DXO/DXO.csproj -c Debug

Alternatively, open DXO.sln in Visual Studio and build the solution.

Run the Application

  dotnet run --project DXO/DXO.csproj

Notes:

Configuration Notes

Server URLs and ports can be configured via:

Security & Authentication

The application now includes comprehensive authentication and authorization:

📖 See Authentication Setup Guide for detailed configuration instructions.

Additional security features:

Visual Walkthrough

Open the application. If running locally, go to http://localhost:5000. This is the home page where you will begin your experience. Diagram

Login with your provider of choice to go to the main page. Alternatively you can use the local development auth bypass option to skip this step.

Diagram

Step 1: Configure Models

Before you create agents, you need to be able to configure the app with your models and provide the endpoint and API Key info. These details will be used when the model is invoked by the agent.

Diagram

Step 2: Configure the Agent Council

Now it’s time to configure your reviewers. These are personas that will evaluate the generated content and provide actionable input to the Creator agent. By default the app comes configured with a default reviewer which you can customize to your liking.

Diagram

Additionally, the app features a gallery of reviewer templates that are designed to excel at specific tasks like data analytics, Code appropriateness, security controls etc. You can add these reviewer personas by clicking the Add reviewer button.

Diagram

You can also customize them by modifying the root prompt, once they are added.

Note: In theory, you can add unlimited number of reviewers. However, beware that, just like in real life, more reviewers cause more churn towards converging the topic. So be mindful of adding more than 3-4 reviewers.

Diagram

Step 3: Define a topic and Start a Session

Diagram

Diagram

Additional Capabilities

Diagram

This feature enables audit-friendly tracking of the entire decision-making process, allowing you to provide input on specific iterations and review the complete history of the Council’s work.

Troubleshooting

Configuration Load Errors

External Model API Errors

SignalR Connection Issues

How It Works (Conceptual Architecture)

  1. Draft Creation
    The Creator agent generates an initial draft from the provided topic.

  2. Parallel Review
    The draft is sent to all configured reviewer agents for critique.

  3. Iterative Refinement
    Reviewer feedback is consolidated and fed back to the Creator. This loop continues until all reviewers sign off.

  4. Final Output
    Once consensus is reached, the final version is produced.

  User Prompt
      |
      v
  +----------------+
  |     Creator     |
  | (Council Chair) |←---------------------
  +----------------+                      |
        |                                 |
        v                                 |
  +-----------------------------+         |
  | Parallel AI Reviewers       |         |
  |  - Technical Reviewer       |         |
  |  - Policy Reviewer          |         |
  |  - Executive Reviewer       |         |
  +-----------------------------+         |
        |                                 |
        v                                 |
  +-------------+                         |
  | Synthesis   |                         |
  | Threshold   |--------------------------
  | Evaluation  |
  +-------------+
        |
        v
  Final Output

The Creator agent determines:

Example Use Cases

Planned enhancements:

Contributing

Contributions are welcome. Ways to contribute:

Please open an issue or submit a pull request.

License

This project is licensed under the MIT License. See the LICENSE file for details. Link to repo: https://github.com/shriramnat/AI-Decision-Council/

This site is open source. Improve this page.