Aivora – AI Platform Script: AI Chat with MCP Tools, Website Builder, Writing Templates & Image Generation on One Credit Balance (Laravel 13)

Aivora - your own AI platform, ready to earn

Aivora is a complete AI platform for one website. You connect your own API keys for OpenAI, Anthropic, Google Gemini and Stability AI; your members buy a plan period or a credit pack and spend one credit balance on AI Chat (files, folders, share links, streaming, MCP tools), a website builder with a sandboxed live preview and ZIP download, 16 writing templates with DOCX and PDF export, and image generation. Credits are charged on the provider's real token usage, failed calls are refunded, and the admin dashboard shows revenue against the estimated provider cost.

Built on Laravel 13 (PHP 8.3+) with a browser installer - the vendor folder is included, no Composer, Node.js or SSH needed. Light theme by default with a full dark theme; every colour can be changed in the admin. One installation is one website.

Every demo below opens with one click - no typing needed. The live demo uses a built-in demo provider, so its answers are labelled sample replies.

• Demo website URL: https://aivora.anowhosting.com/
• One-click member demo URL (Jordan Lee, Pro: chats, website project, documents, images): https://aivora.anowhosting.com/demo-login/member?to=/chat
• One-click admin (owner) demo URL: https://aivora.anowhosting.com/demo-login/owner
• One-click admin demo URL: https://aivora.anowhosting.com/demo-login/admin?to=/admin/mcp
• One-click support demo URL: https://aivora.anowhosting.com/demo-login/support
• Demo overview page: https://aivora.anowhosting.com/codester/index.html
• Public documentation: https://aivora.anowhosting.com/codester/docs/index.html
• Member demo login (email / password): member@aivora.test / password
• Owner demo login (email / password): owner@aivora.test / password
• Team demo logins (email / password): admin@aivora.test, support@aivora.test / password

AI chat with files and tools

• A model picker with prices: GPT-4o, GPT-4o mini, Claude Sonnet, Haiku and Opus, Gemini 2.5 Pro and Flash - only the models of the member's plan
• Conversations with folders and search, PDF / text / image upload where the model supports it, streaming answers with Stop, Regenerate and code blocks with Copy
• Read-only share links (30 days), credits and tokens on every answer
• Remote MCP servers give the chat real tools, shown as "Used tool" cards; write tools ask for confirmation

A website builder with a safe preview

Project type, sections, style, colours, fonts, language, framework and photos go in; a validated file set comes out. The build runs in the background and is shown live - files appear while they are written, with a code preview and a Stop button, and it keeps running when the member leaves the page. Members browse the file tree, read the code, preview the site in a sandboxed frame (generated code never runs on your server), refine it with a prompt, restore earlier versions and download a ZIP.

Writing templates with exports

Sixteen templates - blog post, product description, email, newsletter, ad copy, social post, SEO meta, resume, cover letter, summarise, rewrite, translate and more - with fields, tone and length. The result opens in an editor and exports to DOCX, PDF, Markdown, TXT and HTML. Templates are editable and you can add your own.

Image generation

Prompt, style presets, five sizes and up to four images with GPT Image, Imagen 3 or Stable Image Core; variations, history and downloads as PNG, JPG or WebP. Images are private to the member and kept for 90 days (a setting).

Credits that are fair and clear

• An estimate before sending and the real cost after, in tenths of a credit
• Credit price per model per 1K tokens or per image, set by you; failed provider calls refunded automatically
• Plan credits roll over on Pro and Business; packs valid 12 months; 100 free credits at sign-up
• Dashboard with credits left, usage by feature and credits per day

Plans, packs and six ways to pay

• Free, Starter, Pro and Business for a month or a year - no automatic renewal, a reminder with a pay link instead - plus credit packs; everything editable
• Stripe, PayPal, Paystack, PayHere, Razorpay and bank transfer with receipt upload - every payment verified on the server, signed fail-closed webhooks
• US sales tax by state, invoices, refunds recorded; credits are added only after a verified payment

Your keys, your prices, your tools

Encrypted API keys with a Test button, the credit price of every model with the estimated provider cost and margin, a demo provider for a free public demo, and remote MCP servers over Streamable HTTP with tools on / off, per-plan access and a call log.

One admin panel for the team

A dashboard with users, active plans, revenue, credits used, estimated provider cost and margin from real data; users with usage, ledger and manual credits; payments with bank transfers to approve and sales tax rates; writing templates and builder types; settings, website content, AdSense, SEO, languages with auto-translate, message templates, backups, maintenance and update upload. Owner, admin and support roles with per-person sections and an activity log.

What you get

• The complete script with the vendor folder and browser installer (MySQL / MariaDB or SQLite)
• Optional demo data: Hearth & Honey Bakery in Portland with chats, a website project, documents and images, about 8,400 members, twelve months of payments and usage and four MCP servers with call logs
• Illustrated HTML documentation with screenshots of both themes

Good to know

• You bring your own provider API keys and pay the providers for what your members use; you set the credit prices. The margin shown is an estimate from the provider prices you enter.
• Each provider key has a built-in Test button that checks the key and every enabled model.
• MCP: remote servers over Streamable HTTP; no local (stdio) servers. One admin credential per server - members do not connect their own.
• Plans are period-based: one payment buys one month or one year, and nothing is charged automatically. Online payments need your own merchant accounts.
• Hosts that buffer PHP output get whole answers instead of word-by-word streaming (automatic fallback, or a setting).
• One installation is one website: no team workspaces, no members' own API keys, no publishing of builder sites to hosting. English is bundled; other languages are added in the admin.
• The demo people, businesses, chats and numbers are sample content.

Requirements at a glance

PHP 8.3+, MySQL 5.7+ / MariaDB 10.3+ (or SQLite), Apache, LiteSpeed or Nginx, a cron job, outgoing HTTPS to the AI providers. No Composer, Node.js or SSH needed.

Support and updates

Support covers installation, configuration and bugs in the item. Server administration, provider accounts and custom development are not included.

Demo note: the live demo runs with demo mode on - settings, keys, uploads and deletes are disabled; chatting, building and refining websites, writing and exporting documents, generating images and the checkout up to the gateway (test keys) can be tried with labelled sample answers, and the demo data reloads every 6 hours.

• PHP 8.3 or newer (8.3, 8.4)
• MySQL 5.7+ / 8.x or MariaDB 10.3+ (SQLite 3 for local tests)
• PHP extensions: pdo (pdo_mysql or pdo_sqlite), mbstring, openssl, tokenizer, xml, ctype, json, fileinfo, curl, gd, zip
• Apache with mod_rewrite (cPanel, Plesk, DirectAdmin), LiteSpeed or Nginx; Windows with XAMPP / Laragon
• Outgoing HTTPS from the server to the AI provider APIs; PHP max_execution_time of 120 s or more recommended
• One cron job every minute (Task Scheduler on Windows)
• HTTPS recommended (webhooks, browser notifications, social sign-in)
• No Composer, Node.js or SSH needed - the vendor folder is included
• Accounts you bring: API keys of the AI providers you offer, payment gateway merchant accounts, SMTP; optional Google / Facebook OAuth, AdSense, Google Analytics, WhatsApp / SMS provider, remote MCP servers

Install

1. Create an empty MySQL / MariaDB database and user in your hosting panel and select PHP 8.3 or newer.
2. Upload the aivora folder and point your domain to its public folder - or upload its contents straight into public_html (with the hidden .htaccess files).
3. Open your-domain.com: the browser installer checks the server, asks for the database and creates your site and owner account (demo data optional, imported in short steps so shared hosts do not time out).
4. Sign in at /staff/login, add your provider API keys in Providers & models (Test key), check the credit prices, plans and packs, switch on your payment gateways and set up SMTP.
5. Add the cron job * * * * * cd /path/to/aivora && php artisan schedule:run for monthly credits, expiry, reminders, retention and backups.

Updates

1. Back up, upload the update zip in Admin › Settings › Update (or copy the new files over the old ones; .env, storage and public/uploads are kept), then click Run update - your data is kept.

The illustrated guide (cPanel, VPS / Nginx, Windows and Task Scheduler, the installer, providers and keys, credit prices, streaming, MCP, the builder sandbox, exports, plans and payments with webhooks, languages, demo mode, troubleshooting) is in documentation/index.html and online at /codester/docs/index.html.

1.0.6 – October 2026

• Website builder: Generate and Refine now show the build live, like AI Chat – the files appear while they are written, with a short code preview, the elapsed time and a Stop button. The build keeps running when you leave or reload the page; the project list shows Building… and the result opens when it is done.
• Website builder: a whole website gets enough room – Most tokens for one build is now 32,000 by default (never above the model's own longest reply) and is no longer limited by the chat reply length of the site or the plan. Members still pay only for the tokens really used.
• Website builder: when a site is too large for one answer, the build ends with a clear message ("choose fewer pages or sections, or a simpler style") and is charged once for what was used. If the model's answer cannot be used, it is told exactly why once; code written as text is turned into files. Builds that do not finish can be tried again with one click.
• Website builder: before a build starts the page checks that the member has about the estimated credits and says so if not. The demo provider shows the live build too.
• Website builder: when a model that thinks first uses its whole length limit before writing any file, the build says so clearly instead of failing without a reason.
• Real-model checks for Claude: Test key now recognises every enabled Claude model, also the short model names the provider's list shows only with a date; a "Longest reply" set above a model's own limit is corrected automatically on the next call instead of failing; the known output limits of older Claude models are built in; and when a model that thinks first spends a chat answer's whole limit before writing, the answer says so clearly (writing tools and auto-translate leave room for it too).
• Phones: the live build view is compact in light and dark, the Stop button is always within reach.

1.0.5 – October 2026

• AI Chat: large file requests (a whole website as a ZIP) now get their own, much larger answer length – new setting Most tokens for an answer that creates files in Settings › AI & streaming › Chat (32,000 by default, never above the model's own longest reply). Members still pay only for the tokens really used.
• AI Chat: when files are too large for one answer, the answer ends with a clear note ("ask for fewer or smaller files, or use the Website builder") instead of trying again. If a file request goes wrong twice in one answer, it stops with a clear note too.
• File requests are read more reliably with OpenAI, Anthropic and Gemini (parallel tool calls, thinking models, compatible API addresses), and the model is told exactly which part of a file request was wrong.
• Phones: the file card being written is compact – the files show as chips, the code preview is about eight lines, scrolls inside its own box and can be hidden with Hide / Show code. Long code blocks in answers show their first lines with Show all … lines and scroll inside their own box; nothing is wider than the chat. Light and dark.

1.0.4 – October 2026

• AI Chat shows the writing live all the time: text word by word, and when the model writes files a file card that grows while they are written – ZIP name, each file as soon as its name appears, the size so far and a scrolling code preview of the file being written. Tool calls show as cards while they are prepared.
• Answers keep being written when the member switches chats, reloads or closes the tab; coming back shows the answer still writing or finished, and the chat list marks chats that are still writing. Stop is its own request and works in every streaming mode. One answer runs per chat at a time.
• Hosts that buffer the stream now show the answer's progress too (the page asks the server for it while the answer is written).
• An answer whose server process died is marked failed and refunded automatically. New settings in Settings › AI & streaming › Chat: longest time one answer may run, and when an answer that stopped is marked failed.
• Provider and model icons instead of letter tiles: built-in OpenAI, Claude, Gemini and Stability AI marks in the model picker, chat, pricing, dashboards, images page and admin. Admin › Providers & models: upload your own icon per provider (SVG, PNG, JPG or WebP – SVG files are cleaned) or go back to the built-in one, and give a model its own icon.
• The demo provider writes a small three-file website live when asked for a website or a ZIP.

1.0.3 – October 2026

• AI Chat: Download button on every code block and Download all (ZIP) for answers with several files.
• AI Chat: downloadable files – when a member asks for files, a ZIP or a download, the model creates them and the answer shows a file card with a Download button (also on shared chats). New settings: Chat instructions, Downloadable files in chat with size and file limits; new plan switch "Downloadable files in chat" (on for the paid plans).

1.0.2 – October 2026

• Clearer provider errors for the admin: a provider account without credit (402) and an unknown model name (404) now have their own reasons, the provider's own message is kept with the failed call, and the provider card or the model shows the cause.
• Test key also checks that each switched-on model of the provider exists under its API model name.
• Current Claude model names: Claude Sonnet claude-sonnet-4-6, Claude Haiku claude-haiku-4-5, Claude Opus claude-opus-4-8 (installed sites move over automatically where the starter model name was unchanged).
• Anthropic requests no longer send a temperature setting, so every current Claude model accepts them (auto-translate included).

1.0.1 – October 2026

• Admin › Providers & models: the Replace / Add key button opens the key window on every device again.

1.0.0 – October 2026

• First release: AI Chat with folders, files, streaming, share links and MCP tools; the website builder with a sandboxed preview, versions and ZIP download; 16 writing templates with DOCX, PDF, Markdown, TXT and HTML export; image generation with styles, sizes, variations and three download formats; OpenAI, Anthropic, Google Gemini and Stability AI with encrypted keys and per-model credit prices; credits charged on real usage with refunds for failed calls; plans for a month or a year (no auto-renew) with rollover and credit packs; Stripe, PayPal, Paystack, PayHere, Razorpay and bank transfer; sales tax and invoices; one admin panel with three team roles; light and dark themes with colour presets; languages with auto-translate; browser installer, demo mode, backups and update upload.