Multi-Agent Brainstorming System

A sophisticated, AI-powered brainstorming platform that orchestrates a team of specialized agents to generate, critique, and refine ideas in real-time. The system features continuous debate, progressive depth, intelligent insight clustering, and now a fully customizable agent builder.

🚀 Features

🧠 Intelligent Agents & Builder

The system orchestrates a diverse team of AI personae:

• Core Team: Optimist, Skeptic, Analyst, Evaluator.
• Custom Agents: Create your own agents with unique names, roles, and system prompts.
• Agent Library: Quickly onboard specialized agents (e.g., Growth Hacker, Legal Counsel, Visionary) from our built-in library.
• Flexible Management: Assemble your squad by toggling default agents on/off and removing custom ones.

🔄 Continuous & Progressive Brainstorming

• Infinite Loop: The agents engage in a continuous round-robin debate that runs indefinitely until you decide to stop.
• Progressive Depth: The conversation evolves intelligently over time:
  • Round 0: Creative Idea Generation (Quantity & Novelty)
  • Round 1: Critical Analysis & Risk Identification
  • Round 2: Concrete Solution Proposals & Refinements
  • Round 3+: Deep Debate, Edge Cases, and Long-term Implications

📂 Contextual Intelligence

• File Uploads: Attach PDFs or Images to your brainstorming session. The system uses OCR to extract text and context, allowing agents to analyze your documents directly.
• Semantic Clustering: Automatically groups hundreds of generated ideas based on semantic similarity using embedding models.

💻 Modern Experience

• Real-Time Streaming: Watch agents type out their thoughts live via Server-Sent Events (SSE).
• Glassmorphic UI: A beautiful, dark-mode interface with cinematic animations.
• Full Control: Pause/Resume the stream or End the session at your convenience.

---

🛠️ Technology Stack

• Frontend: Next.js 16, React 19, Tailwind CSS v4
• Backend: FastAPI, Python 3.11+, Server-Sent Events (SSE)
• AI/ML:
  • LLM: Google Gemini 2.5 Flash (via google-generativeai) or Ollama (Local Fallback)
  • Embeddings: Gemini Embedding API (models/embedding-001) or Ollama fallback
  • Clustering: Scikit-Learn (Agglomerative Clustering)
• Database: Supabase (PostgreSQL + pgvector) — optional; in-memory mode works without it

---

⚡ Quick Start

Prerequisites

• Python 3.11+
• Node.js 18+
• Supabase Account (optional — app runs in-memory without it)
• Google Gemini API Key (optional — defaults to Ollama local fallback if missing)

Architecture

The backend is FastAPI (port 8000) — it handles all business logic, LLM calls, database operations, and streaming. Next.js (port 3000) is the frontend that proxies all API requests to FastAPI. FastAPI is the source of truth.

User → Next.js (:3000) → FastAPI (:8000) → LLM/Database

One-Click Setup (Mac/Linux)

We provide a unified startup script that handles virtual environments, dependencies, and server launching.

# 1. Clone the repo
git clone https://github.com/vedantdhoke/multi-agent-brainstorming-system.git
cd multi-agent-brainstorming-system

# 2. Configure Environment
# Create a .env file in the root directory (or let the script warn you)
echo "SUPABASE_URL=your_url" >> .env
echo "SUPABASE_KEY=your_key" >> .env
echo "LLM_API_KEY=your_gemini_key" >> .env
echo "GOOGLE_API_KEY=your_google_key" >> .env

# 3. Launch
./start_dev.sh

Manual Setup

Backend (FastAPI)

cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Frontend (Next.js Proxy)

cd frontend
npm install
npm run dev

The frontend automatically proxies API calls to http://127.0.0.1:8000 (FastAPI). To change this, set the FASTAPI_BASE_URL environment variable.

Usage

1. Open http://localhost:3000.
2. Assemble Your Squad: Select default agents or create new ones from the Library.
3. Set Context: Enter a topic (e.g., "Future of Space Travel") or upload a PDF.
4. Click "Start" and watch the debate unfold.
5. Use the Analysis tab to see clustered insights.

---

🔒 Architecture

1. Orchestration: The Orchestrator service cycles through agents, managing the context window and injecting progressive system prompts based on the round number.
2. State Management: Every response is asynchronously streamed to the client and persisted to the responses table in Supabase.
3. Clustering Pipeline:
   • Fetches text -> Generates Embeddings -> Runs Agglomerative Clustering.
   • Samples text from each cluster -> Prompts LLM for a title -> Saves to clusters table.