VorexO Docs

Introduction

VorexO is a mobile IDE for Android built around a simple idea: your phone is enough to write and ship small projects. It bundles a code editor (CodeMirror 6), a live preview sandbox, a project file manager, and an AI assistant that can explain code, plan features, or write entire files for you.

The AI side follows a BYOK (bring your own key) model. VorexO ships without any API keys and has no backend of its own — you pick a provider, paste your key in Settings, and the app talks to that provider directly from your device. Twelve providers are supported out of the box, including a local Ollama instance running on your own network.

Details

Features

The three AI modes

Every conversation runs in one of three modes, switched from the selector above the input:

Installation

VorexO is distributed as a direct-download APK (VorexO-v1.0.apk). It isn't on Google Play yet, so Android will ask you to allow installs from your browser or file manager once.

Permissions

That's the whole list — no contacts, location, storage-wide access or phone state. The build also disables Android backup (allowBackup=false), so app data isn't copied around during device transfers.

Quick start

From install to a working page, roughly two minutes.

landing-page/
├── index.html   ← entry point
├── style.css
└── script.js

Editor tips

• The toolbar above the keyboard inserts brackets, quotes and tabs so you don't fight the soft keyboard.
• Long-press a file to rename, duplicate or delete it.
• Project menu → Export zips the whole project for sharing or backup.

AI providers

All providers work the same way: create an account, generate an API key, paste it in Settings → AI Providers. Keys are stored locally on your device and sent only to the provider you configured — they're never embedded in the app, never proxied, and (as of v1.0) excluded from cloud sync.

Ollama over your network

The Ollama provider accepts a full base URL, so the model can run on a PC on the same Wi-Fi while you code on the phone. Start Ollama listening on all interfaces, then point VorexO at the machine's LAN IP:

# expose Ollama on your LAN, then pull a model
OLLAMA_HOST=0.0.0.0:11434 ollama serve
ollama pull qwen2.5-coder:7b

# in VorexO: Settings → Ollama → Base URL
http://192.168.1.42:11434   ← your PC's LAN IP, key stays empty

Key safety

• Set a spending limit on the provider's dashboard before pasting a key anywhere.
• Use scoped or capped keys where offered (OpenRouter key limits, OpenAI project keys).
• Cloud sync deliberately excludes API keys — after restoring on a new device, re-enter them manually.

Agent mode

Agent mode changes the system prompt so the model stops describing code and starts shipping files. The model replies with a one-line plan, then emits every file inside this wrapper:

I'll build a modern landing page.

[FILE:index.html]
<full, complete HTML code here — no placeholders>
[/FILE]

[FILE:style.css]
/* full CSS here */
[/FILE]

[FILE:script.js]
// full JS here
[/FILE]

VorexO parses the reply and shows a card per file: path, syntax-highlighted preview, and Save / Copy / Dismiss buttons. Saving writes (or overwrites) that path in the current project — existing files are marked as an update.

Getting good results

• Name your stack — "plain HTML, one CSS file, vanilla JS, no build step" keeps the agent inside what the preview can run.
• Split big tasks. Ask for the file list first, then request files one by one; short replies fail less than a whole project in one shot.
• Mind the output limit. Each model has a maximum response length. v1.0 clamps maxTokens to the model's real cap, but a truncated file can still happen on small models — ask for a re-emit of just that file.

Cloud sync

Sync is optional and runs on your own Firebase project. The app ships with its default project preconfigured, but plugging in your own Firebase web config (Settings → Cloud) gives you full ownership of the data. Sign in with email, and VorexO stores a snapshot of your workspace in Realtime Database.

What gets synced

Each account gets its own node:

vorexo-default-rtdb
└── users/
    └── uid_<firebase-uid>/
        ├── projects/   ← every project and file
        └── settings/   ← app preferences, no keys

Lock the database down

If you use your own Firebase project, replace the default rules so only the owning account can read or write its node. Firebase Console → Realtime Database → Rules:

{
  "rules": {
    "users": {
      "$uid": {
        ".read":  "auth != null && auth.uid === $uid.replace('uid_', '')",
        ".write": "auth != null && auth.uid === $uid.replace('uid_', '')"
      }
    }
  }
}

To restore: fresh install → Settings → Cloud → sign in → Load from Cloud. Projects and settings come back immediately; re-enter your API keys and you're done.

FAQ

Changelog

Streaming responses (token-by-token) are the known gap — the embedded HTTP layer buffers full replies. A native streaming plugin is on the roadmap.