---

Overview

Vectaix AI is a production-grade, multi-model AI chat platform that unifies the world's most powerful language models under a single, elegant interface. Rather than locking users into one AI provider, Vectaix gives you the freedom to switch between — or even combine — frontier models seamlessly.

At its core is Fusion Mode, a novel multi-agent consensus framework that dispatches queries to multiple frontier LLMs in parallel and synthesizes their outputs through structured deliberation — dramatically reducing hallucination and bias.

---

Features

🤖 Multi-Model Intelligence

Access 6 direct chat models plus Fusion Mode through a unified interface. Switch models mid-conversation with full context preservation.

Model	Provider	Context Window	Inputs	Thinking	Web Search
Fusion	OpenRouter	—	Text	—	—
Auto Router (Balanced)	AIHubMix	Dynamic	Text, Image	✅	✅
GPT-5.6 Sol	OpenAI	1.05M	Text, Image	✅	✅
Claude Opus 4.8	Anthropic	1M	Text, Image	✅	✅
Gemini 3.6 Flash	Google	1M	Text, Image, Audio, Video	✅	✅
Grok 4.5	xAI	500K	Text, Image	✅	✅
Kimi K3	Moonshot AI	1M	Text, Image	✅	✅

Dedicated media models:

Model	Provider	Capability
GPT Image 2	OpenAI	Image generation and image editing
Seedance 2.0 Standard	ByteDance	Text-to-video and image-to-video

🏛️ Fusion Mode — Multi-Agent Consensus

The crown jewel of Vectaix AI. Inspired by the deliberative processes of real-world councils, this mode orchestrates multiple AI experts to arrive at a more truthful, balanced answer.

                              ┌─────────────────┐
                              │   User Query     │
                              └────────┬─────────┘
                                       │
                          ┌────────────┼────────────┐
                          ▼            ▼            ▼
                   ┌────────────┐┌────────────┐┌────────────┐
                   │    GPT     ││   Claude   ││   Gemini   │
                   │    5.5     ││  Opus 4.8  ││ 3.6 Flash  │
                   │  (Expert)  ││  (Expert)  ││  (Expert)  │
                   └─────┬──────┘└─────┬──────┘└─────┬──────┘
                         │             │             │
                         └─────────────┼─────────────┘
                                       ▼
                              ┌─────────────────┐
                              │OpenRouter Fusion│
                              │  (Synthesis)    │
                              └────────┬─────────┘
                                       │
                         ┌─────────────┼─────────────┐
                         ▼             ▼             ▼
                   ┌──────────┐ ┌──────────┐ ┌──────────┐
                   │Agreement │ │Key Diffs │ │ Unique   │
                   │  Points  │ │& Debates │ │ Insights │
                   └──────────┘ └──────────┘ └──────────┘

How it works:

1. Parallel Generation — Your query is simultaneously sent to GPT-5.6 Sol, Claude Opus 4.8, and Gemini 3.6 Flash
2. Independent Reasoning — Each expert reasons independently with its own strengths and knowledge
3. Structured Synthesis — OpenRouter Fusion analyzes all responses, identifying:
   • ✅ Agreement — Points where all experts converge
   • ⚖️ Key Differences — Where experts disagree and why
   • 💡 Unique Insights — Valuable perspectives from individual experts
   • 🔍 Blind Spots — Gaps that only cross-model analysis reveals

Fusion Mode currently does not support web search; it focuses on text-only multi-model reasoning.

Key Results from Research:

Benchmark	Improvement
HaluEval (Hallucination Detection)	35.9% relative reduction
TruthfulQA	+7.8 points over best individual model
Cross-domain Bias Variance	Significantly lower

🌐 Web Browsing & Search

Real-time access to the internet with intelligent multi-round browsing capabilities.

• Smart Search — Discover fresh, relevant sources with Firecrawl Search
• Page Crawling — Extract selected pages as clean Markdown with Firecrawl Scrape
• Multi-Page Browsing — Crawl multiple pages in a single session
• Inline Citations — Every claim backed by traceable source references

📎 Private Multimodal Files

Upload private media files stored on the Zeabur mounted disk.

File Type	Supported Formats	Capability
🖼️ Images	PNG, JPG, GIF, WebP	Visual analysis, OCR, description
🎵 Audio	MP3, WAV, M4A, AAC, OGG	Gemini audio understanding
🎬 Video	MP4, MOV, WebM, M4V	Gemini video understanding

✨ Polished User Experience

💬 Conversation Management Persistent chat history with MongoDB Intelligent long-conversation compression Pin important conversations Conversation-specific model & settings	🎨 Themes & Customization Dark / Light mode with smooth transitions Adjustable font size Completion sound with volume control Custom user avatars
📝 Rich Markdown Rendering Full GitHub Flavored Markdown (GFM) LaTeX math equations (KaTeX) Syntax-highlighted code blocks Scrollable tables with copy support	🔐 Authentication & Security MongoDB-backed server session management Bcrypt password hashing Rate limiting on all endpoints Admin user management panel
⚙️ Advanced Controls Per-model thinking level adjustment Max tokens control Custom system prompts with presets Media resolution settings	📱 Progressive Web App Installable on any device Mobile-optimized responsive UI Touch-friendly interface Offline-capable manifest

---

Architecture

vectaix-ai/
├── app/
│   ├── api/
│   │   ├── fusion/           # Fusion Mode orchestration
│   │   ├── chat/             # Multi-provider chat & compression
│   │   ├── auth/             # Authentication endpoints
│   │   ├── conversations/    # Conversation CRUD
│   │   ├── media/            # Image/video generation
│   │   ├── upload/           # Private disk file upload
│   │   └── admin/            # Admin management
│   ├── components/           # React UI components
│   │   ├── chat/             # Chat input & model selector
│   │   ├── message/          # Message display components
│   │   │   ├── FusionMessage.js # Fusion Mode result rendering
│   │   │   ├── MessageList.js
│   │   │   └── ...
│   └── ChatApp.js            # Root application component
├── lib/
│   ├── client/               # Client-side utilities
│   │   ├── chat/             # Chat actions & runtime
│   │   └── hooks/            # React hooks (theme, settings)
│   ├── server/               # Server-side logic
│   │   ├── chat/             # Provider adapters, config, prompts
│   │   ├── webBrowsing/      # Web search & crawl engine
│   │   ├── storage/          # Mounted-disk storage service
│   │   └── conversations/    # Conversation storage logic
│   └── shared/               # Shared constants & types
│       ├── models.js         # Model definitions & capabilities
│       ├── attachments.js    # File type handling
│       └── webSearch.js      # Search configuration
├── models/                   # Mongoose schemas
│   ├── User.js
│   ├── Conversation.js
│   └── StoredFile.js
└── public/                   # Static assets

---

Tech Stack

Next.js 16

React 19

Tailwind CSS

MongoDB

Node.js

Layer	Technologies
Frontend	Next.js 16 · React 19 · Tailwind CSS · Framer Motion · Ant Design · Lucide Icons
Backend	Next.js API Routes · Node.js · SSE (Server-Sent Events) Streaming
Database	MongoDB with Mongoose ODM
Storage	Zeabur mounted disk (private media files)
AI Providers	Inferera · Google Gemini API · OpenRouter
Auth	MongoDB server sessions · bcryptjs
Rendering	react-markdown · rehype-highlight · rehype-katex · remark-gfm · remark-math
Deployment	Zeabur native Next.js service (single instance)

---

Getting Started

Prerequisites

• Zeabur project connected to the Git repository
• MongoDB instance
• API keys for at least one AI provider

Deploy on Zeabur

1. Create a service from the Git repository with Deploy your source code. Zeabur will automatically recognize the Next.js project and run its npm build/start scripts.
2. Create a persistent volume named vectaix-data and mount it at /data.
3. Set the HTTP port to 3000 and the health check path to /api/health.
4. Add the environment variables below, then redeploy.

Environment Variables

Variable	Required	Description
MONGO_URI	✅	MongoDB connection string
STORAGE_ROOT	✅	Mounted disk directory, use /data/vectaix on Zeabur
ADMIN_EMAILS	—	Comma-separated administrator email addresses
AIHUBMIX_API_KEY	—	Shared API key for AIHubMix Auto Router, Kimi K3 via Anthropic Messages, and Inferera GPT, Grok, Claude, image, and video APIs
GEMINI_API_KEY	—	Google Gemini API key for Gemini chat and conversation compression
OPENROUTER_API_KEY	—	OpenRouter API key for Fusion synthesis
FIRECRAWL_API_KEY	—	API key for Firecrawl Search and Scrape

[!TIP] Configure the key for every model or feature you enable. Missing keys return a clear configuration error.

---

License

This project is licensed under the MIT License.

---