TL;DR: We’re open-sourcing a complete, production-ready UI StarterKit for CustomGPT.ai’s RAG API. MIT licensed. Voice mode included. Deploy anywhere. Try the instant demo. Or clone it on GitHub.
See the live embed below!
The Problem We All Know Too Well
If you’re using CustomGPT.ai’s API, you know the drill. The API is solid – great RAG capabilities, reliable performance, decent pricing. But when it comes to the UI, it gets all the boxes checked but you don’t get a 100% customisability, but now you DO.
Your clients want custom branding. Your product team wants voice interactions. Your compliance team wants to enforce rate limits that the API doesn’t natively support. And you? You just want to ship something that doesn’t look like it was built in 2019.
We’ve been there. We’ve built the same proxy handlers, streaming parsers, and state management solutions over and over.
Today, we’re done rebuilding the wheel, and so are you.
Introducing the CustomGPT.ai Developer Starter Kit
We’re releasing a new complete frontend implementation for CustomGPT.ai as open source software that you can use as a starter kit (think ChatGPT interface but you can embed this on any website, add floating chatbot icon that too all multi-instance). Not “open source with strings attached” – proper MIT licensed code that you can use, modify, and deploy however you want.
🚀 Live Demo: starterkit.customgpt.ai
📦 GitHub: github.com/Poll-The-People/customgpt-starter-kit
📚 Documentation: https://docs.customgpt.ai/reference/full-fledged-chat-ui-with-project-settings
What’s In The Box?
Core Features That Actually Matter
1. Complete API Coverage Every single CustomGPT.ai endpoint is wrapped and ready. Messages, conversations, projects (yes, the API calls them “projects” while the UI says “agents” – we handle that confusion for you), citations, file uploads – it’s all there.
2. Voice Mode That Just Works
- 6 different voices powered by OpenAI’s TTS
- Whisper integration for speech-to-text
- Streaming audio responses
- Under 100ms latency on edge runtime
- Optional: only needs OPENAI_API_KEY if you want voice features
// Your users can literally just talk to your RAG
const response = await transcribeAudio(audioBlob);
const aiSpeech = await generateSpeech(response.text);
3. Multiple Deployment Modes
Deploy the same codebase in three different ways:
<!-- Option 1: Embedded Widget (there are 2 widgets) -->
<script src="https://your-domain.com/widget.js"></script>
<script>
CustomGPTWidget.init({
agentId: 'your-agent-id',
theme: 'dark',
position: 'bottom-right'
});
</script>
<!-- Option 2: Iframe -->
<iframe src="https://your-domain.com/chat/agent-id" />
<!-- Option 3: Full Standalone App -->
<!-- Deploy the entire Next.js app -->
4. Security-First Architecture
Your API keys never touch the client (they are safely stored in your .env files):
// Old way (exposes your key)
fetch('https://api.customgpt.ai/v1/messages')
// Our way (proxy through your server)
fetch('/api/proxy/messages')
5. Real-Time Streaming That Doesn’t Leak Memory
We spent way too much time getting SSE streaming right:
// Proper cleanup, error boundaries, backpressure handling
const stream = new EventSource('/api/proxy/stream');
stream.onmessage = (e) => {
// Automatic cleanup on unmount
// Memory leak prevention
// Error recovery built-in
};
The Technical Deep Dive
Architecture That Scales
Built on Next.js 14 with TypeScript throughout. State management via Zustand with persistence. Styling with Tailwind CSS (dark mode + full mobile responsiveness included, obviously).
src/
├── app/api/proxy/ # Secure API proxy handlers
├── components/chat/ # Modular chat components
├── lib/streaming/ # SSE stream management
├── store/ # Zustand state stores
└── widget/ # Standalone widget bundle
Progressive Web App Ready
- Installable: Add to home screen on mobile/desktop
- Offline capable: Service worker included
- App-like: Fullscreen, custom icons, splash screens
Developer Experience First
# Development
npm run dev
# Build widget only
npm run build:widget
# Deploy to Vercel
vercel
# Deploy to Railway
railway up
# Even Google Apps Script (don't ask)
Just get the code there and deploy (for certain social media AI bots)
Enforce Business Logic UI-Side
Things you can now do that the API doesn’t support:
- Limit number of queries per user
- Restrict conversation history
- Disable features per deployment (we got trail mode and demo mode for you to try this out, check at the top)
- Custom rate limiting rules
- UI-level access control
Getting Started in 3 Minutes
Step 1: Clone and Configure
git clone https://github.com/Poll-The-People/customgpt-starter-kit
cd customgpt-starter-kit
cp .env.example .env.local
Step 2: Add Your Keys
# Required
CUSTOMGPT_API_KEY=your_api_key_here
# Optional - for voice features
OPENAI_API_KEY=your_openai_key_here
Step 3: Run
npm install
npm run dev
That’s it. You’re running a custom UI for your CustomGPT agent.
Deploy to Production
Vercel (Recommended): Use this magic link to deploy to Vercel instantly
But Wait, There’s More: Social Bot Integrations
We’re also releasing 9 chatbot integrations that connect your CustomGPT.ai API endpoints agents to social platforms, full ready to use:
📦 GitHub: github.com/Poll-The-People/customgpt-integrations
Supported Platforms:
- Slack – App or webhook mode
- Discord – Full bot with slash commands
- Telegram – Webhook or polling
- WhatsApp – Via Twilio
- Facebook Messenger – Graph API integration
- Instagram – Direct messages
- MS Teams – Bot framework
- Rocket.Chat – REST API
- Google Chat – Apps Script or webhook
Each bot includes:
- Full deployment guides
- Rate limiting
- Error handling
- Conversation context
- Multiple and Free hosting options
Highlight: Several bots can run entirely on Google Apps Script (free tier: 20k requests/day) with just a single file deployment.
Demo Mode: Try Before You Deploy
We built a unique 2-tier demo system:
Free Trial (10 minutes):
- Uses our API key
- Read-only sandbox
- No setup required
- Auto-expires to prevent abuse
Demo Mode:
- Add your own API key
- Stored locally in browser only
- Full functionality
- Key destroyed on tab close
Try it live: starterkit.customgpt.ai
What This Means for the Community
For Developers
- Stop rebuilding the same UI
- Focus on your unique value proposition
- Ship faster with production-ready code
For Businesses
- White-label your RAG solutions
- Enforce business rules UI-side
- Deploy anywhere (on-prem included)
For the Ecosystem
- More innovation on top of CustomGPT
- Better end-user experiences
- Shared knowledge and improvements
Roadmap: What’s Next?
We’re committed to maintaining and improving this starter kit. Coming soon:
- Analytics dashboard for usage tracking
- Plugin system for custom extensions
- Bring in your request, we code it for you.
Join the Movement
This is more than code – it’s a statement. We believe in:
- Open source that’s actually open
- Developer tools that respect developers
- No vendor lock-in, ever
- Community-driven development
Get Involved
⭐ Star the repo: GitHub
🐛 Report issues: GitHub Issues
💡 Contribute: PRs welcome (please)
💬 Join discussion: GitHub Discussions
Quick Links
- 🚀 Live Demo: starterkit.customgpt.ai
- 📦 Starter Kit: github.com/Poll-The-People/customgpt-starter-kit
- 🤖 Bot Integrations: github.com/Poll-The-People/customgpt-integrations
- 📚 Documentation: docs.customgpt.ai
- 🔄 Vercel Template: Deploy with one click
The Bottom Line
We built this because we needed it. We’re open-sourcing it because you need it too.
No strings attached. No premium tiers. No telemetry. Just good code that solves real problems.
Welcome to the future of RAG interfaces. Let’s build something amazing together.
PS: Yes, it runs on Google Apps Script. No, we’re not proud of it. But sometimes you need free hosting that just works, and we’ve got you covered.
About CustomGPT.ai: CustomGPT is a leading RAG (Retrieval-Augmented Generation) platform that enables businesses to build AI agents powered by their own data.
License: MIT – Use it commercially, modify it, sell it. We don’t care. We just want better UIs in the world.
Built with ❤️ by developers, for developers.
Priyansh is Developer Relations Advocate who loves technology, writer about them, creates deeply researched content about them.