Remix

Make your own version of Gondal AI

Get an exact, working copy in your account, then change anything with chat.

Original by M Qasim ZAIN

What would you change?

Start free with 40 credits. No card needed.

Start fresh from this prompt

Builds a brand new app from the first prompt only, so later changes to the original are not included.

# Build Gondal AI — Gondal Family's Private AI Platform

Build a professional, mobile-first, installable Progressive Web App (PWA) named **Gondal AI**, featuring our personal AI assistant **Gondel Nova**.

This is a private AI project for my family, not a public AI SaaS product. Build a real application with working functionality, clean architecture, secure access controls, and a beautiful ChatGPT-quality interface.

## 1. Core mission

Create a family AI app that can be opened through one shareable HTTPS URL on Android phones, iPhones, tablets, and computers.

The app should remain accessible when my personal PC is switched off. Do not make the website, login, chat interface, or AI inference depend on my home PC being online.

Prioritize offline operation, zero per-message AI API charges, user privacy, and ownership of the source code.

## 2. Critical AI architecture

Do not use OpenAI, Anthropic, Gemini, or any paid hosted LLM API. Do not create API keys for paid AI providers or require a subscription for normal model inference.

Support two AI inference modes:

**Mode A — On-device AI**
- Run a compatible open-weight language model directly on the user's device using a browser-based inference engine such as WebLLM/WebGPU, where supported.
- Investigate compatible small Qwen-family models and their actual runtime requirements.
- Download model files only after the user approves, show download progress, and cache supported assets for later offline use.
- Check WebGPU, memory, storage, browser compatibility, and model availability before starting.
- Keep prompts and conversations on the device by default.
- If a device cannot run the selected model, explain the limitation instead of pretending inference works.

**Mode B — Self-hosted Ollama**
- Allow an administrator to configure an Ollama-compatible endpoint running a Qwen model, such as Qwen3.
- Keep endpoint settings configurable and never hardcode secrets into frontend code.
- Explain that this mode requires its configured server to be online.
- Do not silently switch to a paid cloud model if the local server becomes unavailable.

Never claim that a model is downloaded, trained, offline-ready, or working unless the app verifies it.

## 3. Honest offline behavior

The app shell, navigation, cached conversations, settings, and previously saved data should remain available offline after installation and initial setup.

AI chat must work offline only when a compatible model has actually been downloaded and initialized on that device.

Show clear states:
- Online and ready
- Downloading model
- Loading local model
- Ready for offline chat
- Device not supported
- Model unavailable
- Offline — local model required

Do not fabricate responses or silently send private conversations to external services.

## 4. Gondal Family access

Create a private family access flow:
1. Welcome screen with Gondal AI branding.
2. Ask the visitor's name.
3. Show a separate family access-code screen.
4. Validate access on a secure backend, not only through frontend JavaScript.
5. Include a sign-out option, session expiry, and a safe device-remember feature.
6. Include an owner-controlled family member list and the ability to revoke access.
7. Do not expose family codes, password hashes, private model files, or administrator secrets in public frontend code.
8. Provide a safe account-recovery request flow without displaying secrets or sending email automatically unless a real email service is configured.

Only authorized Gondal family members should access protected family resources.

## 5. Premium-quality interface

Design a polished, responsive, dark-first interface with optional light mode.

Include:
- Left sidebar with Gondal AI logo and Gondel Nova branding
- New chat and searchable conversation history
- Rename, archive, pin, and delete conversations
- Main chat area with readable message bubbles
- Streaming-style response display when supported
- Stop generation, regenerate, copy, and retry
- Markdown, code blocks, tables, and formatted answers
- Message composer with attachment controls
- Clear loading, error, and offline indicators
- Model selector showing only models actually available
- Settings for privacy, local model, storage, and theme
- Responsive mobile navigation and touch-friendly controls
- Installable PWA manifest, icons, and service worker
- Accessible controls, keyboard support, and reduced-motion support

Use original Gondal AI branding. Do not copy ChatGPT logos or claim to be OpenAI.

## 6. Chat storage and privacy

- Save local conversations in IndexedDB.
- Clearly distinguish device-local data from any synchronized family data.
- Never synchronize private conversations by default.
- Do not store access codes or passwords in localStorage.
- Add export and import for conversations in a documented format.
- Add a confirmation step before deleting all chats.
- Provide a clear way to remove locally downloaded model files and cached data.
- Minimize external network requests and document every external dependency.

## 7. Model management

Create a working model-management screen that displays:
- Model name and family
- Download status and storage size when known
- Device compatibility
- Model loading status
- Selected inference engine
- Whether offline inference has been verified
- Delete and change-model controls

Use compatible Qwen-family models where technically supported. Do not promise that every Qwen model can run in every mobile browser.

## 8. Owner dashboard

Create an administrator-only dashboard for:
- Family access management
- Model and inference settings
- Privacy and storage settings
- App health and diagnostic information
- Exporting application data
- Revoking sessions and remembered devices

Enforce administrator permissions on the backend wherever protected server operations are involved.

## 9. Technical requirements

Build a real PWA with a maintainable component structure, a clear separation between UI and inference engines, and documented configuration.

Prefer open-source, locally executable inference libraries and models whose licenses permit the intended use.

Do not add paid AI integrations, metered LLM calls, fake offline functionality, hardcoded secrets, or mock chat responses as if they were real AI.

Do not claim the app can perform server-side Qwen inference while all compute servers are switched off. Instead, implement and verify on-device inference for compatible devices and explain the requirements honestly.

If a requested feature cannot be implemented by the platform, clearly identify the limitation and provide the source code and integration point needed to implement it later.

## 10. Acceptance tests

Before calling the app complete, verify:
1. The website opens using its deployed HTTPS URL on a separate device.
2. Unauthorized users cannot access protected family resources.
3. A supported device can download and initialize a compatible local model.
4. After setup, airplane mode still permits local AI chat when the model and required runtime assets are cached.
5. An unsupported device receives an honest compatibility message.
6. Switching off the home PC does not break the deployed website or supported on-device AI mode.
7. No paid LLM API is called.
8. Existing conversations persist locally after restarting the app.
9. Model loading, download failures, and inference errors are handled clearly.
10. The project can be exported as source code with setup and deployment instructions.

Build the working application, not just a visual mockup. Prioritize the core chat, on-device inference, PWA installation, and secure family access before optional features.

Start fresh from this prompt

Builds a brand new app from the first prompt only, so later changes to the original are not included.