Streaming AI Chat
Responses stream into the interface in real time, keeping longer coding and reasoning workflows responsive.
Multimodal AI Platform
Codarox AI One interface. Multiple AI capabilities.
A cloud-powered AI assistant built for conversations, coding, document and image analysis, adaptive model routing and AI image generation in one responsive web experience.
Overview
A practical AI assistant designed as a real product.
Codarox AI combines multimodal chat, coding assistance, image understanding, document analysis and image generation behind a focused interface. The application routes requests through cloud AI providers, streams responses in real time and keeps conversation history available locally without requiring heavyweight inference hardware on the production server.
Features
Built around useful multimodal AI workflows.
Adaptive Model Routing
Auto mode can route requests through Google Gemini and fall back to OpenRouter while manual model selection remains available.
Documents & Images
PDFs, text and code files, plus image attachments, can be added directly to conversations for contextual analysis.
AI Image Generation
A dedicated Create image mode uses Cloudflare Workers AI with FLUX.1 Schnell and supports download and regeneration.
Persistent Conversations
Conversation metadata is stored locally while generated images use IndexedDB so image history survives navigation and refreshes.
Responsive Interface
The product is designed as a focused desktop and mobile web application with adaptive navigation and touch-friendly controls.
Architecture
Thin application server, cloud AI where it matters.
React Client
The Vite-powered React interface manages chat state, attachments, history, responsive UI and generation controls.
Express API
A Node.js and Express server keeps provider credentials private and exposes dedicated chat, title and image-generation endpoints.
Model Router
Requests can use automatic routing or explicit model selection across Google Gemini and OpenRouter-backed models.
Image Pipeline
Cloudflare Workers AI runs FLUX.1 Schnell for image generation while the client persists generated image data in IndexedDB.
Technology
Built with a focused web and AI stack.
React
Vite
Node.js
Express
Google Gemini
OpenRouter
Cloudflare Workers AI
FLUX.1 Schnell
IndexedDB
PDF.js
Streaming APIs
FAQ
Engineering decisions behind Codarox AI.
Codarox AI is a multimodal web assistant that combines AI chat, coding support, document and image analysis, model routing and image generation in a single product interface.
The provider layer avoids coupling the product to a single model vendor. Automatic routing can prefer Gemini and use OpenRouter as a fallback, while explicit model choices remain available for different workloads.
The composer accepts supported images, PDFs and text or code files. They are normalized into the conversation workflow so multimodal models can answer with the attached context.
Create image mode sends a text prompt through the server to Cloudflare Workers AI, where FLUX.1 Schnell generates the image. The result can be downloaded or regenerated from the conversation.
Inference runs through cloud AI providers, so the production server focuses on API routing, credential protection and application delivery instead of requiring expensive local GPU hardware.
Conversation metadata remains lightweight in local storage, while generated image data is stored separately in IndexedDB and restored when the conversation is reopened.
The project demonstrates frontend product design, Node.js API architecture, provider abstraction, multimodal workflows, local persistence and deployment of an AI application as part of the wider Codarox ecosystem.