SynapseForge — Multi-Model Collaborative Intelligence Engine

Personal Project by Samrisht

SynapseForge fuses multiple AI models into a single collaborative intelligence engine. Unlike debate-style systems where models argue against each other, SynapseForge makes models work together — each contributing unique strengths to produce answers that are demonstrably superior to any single model.

🧠 Core Concept

Traditional multi-model systems pit AI models against each other in debate. SynapseForge takes a fundamentally different approach:

• Collaborative, not competitive — Models complement each other's knowledge
• Specialized roles — Contributors, Verifiers, Stress-Testers, and a final Synthesizer
• Cross-validation — Claims are verified across multiple models for reliability
• Consensus detection — Early convergence stops unnecessary rounds, saving cost
• Provider-agnostic — Integrate OpenAI, Google Gemini, Anthropic Claude, or any model with an API key

⚡ Quick Start

Web Interface (Recommended)

pip install -r requirements.txt
python server.py

Then open http://localhost:5000 in your browser.

Streamlit Interface (Legacy)

streamlit run app.py

🔑 API Key Setup

Enter your API keys directly in the web interface, or create a .env file:

OPENAI_API_KEY=sk-...
GOOGLE_API_KEY=AIza...
ANTHROPIC_API_KEY=sk-ant-...

🎭 Demo Mode

Don't have API keys? Select the Demo preset to use mock agents for a zero-cost demonstration of the platform.

🏗️ Architecture

SynapseForge/
├── server.py                    # Flask web server
├── app.py                       # Streamlit interface (legacy)
├── templates/index.html         # Modern web UI
├── static/
│   ├── styles.css               # Design system
│   └── app.js                   # Frontend logic
├── debate_app/
│   ├── agents/providers.py      # Multi-provider model agents
│   └── core/
│       ├── base.py              # Agent base classes
│       ├── prompts.py           # Collaborative system prompts
│       └── pricing.py           # Cost estimation
└── requirements.txt

🤖 Supported Models

Provider	Models	Roles
OpenAI	GPT-4o, GPT-4o mini, GPT-3.5 Turbo	Contributor, Judge, Verifier, Stress-Tester
Google	Gemini 1.5 Pro, Gemini 1.5 Flash	Contributor, Judge, Verifier, Stress-Tester
Anthropic	Claude 3 Opus, Claude 3 Haiku	Contributor, Judge, Verifier, Stress-Tester
Mock	Various mock agents	All roles (free demo)

📊 Features

• Studio — Configure your model team, set budgets, and run collaborative synthesis
• Synthesis Feed — Real-time round-by-round transcript of model collaboration
• Analytics — Cost tracking, token usage, consensus trajectory, provider cost breakdown
• Preset Configs — Balanced, Rigorous, or Demo mode with one click
• Export — Download full run data as JSON

🔬 How It Works

1. You ask a question — Complex research queries work best
2. Models collaborate — Each model independently analyzes the question, then refines based on others' contributions
3. Verification — Optional verification agent cross-checks factual claims
4. Stress-testing — Optional adversarial agent tests edge cases
5. Synthesis — A judge model fuses all contributions into a single, superior answer
6. Output — You receive an answer that leverages the collective intelligence of multiple models

📈 Why Multi-Model Collaboration?

• Reduces hallucination — Cross-validation catches errors single models miss
• Broader knowledge — Different models have different training data strengths
• Higher reliability — Consensus-based answers are more trustworthy
• Cost-efficient — Early convergence detection prevents unnecessary API calls

---

Built for User· 2026