I built an open-source ChatGPT alternative where the "Open" isn't just marketing
TL;DR: GptOpen is an MIT-licensed AI chat app where you create custom AI agents (name, avatar, instructions, personality), chat solo or in group chats, and plug in ANY API model — Groq, OpenRouter, Gemini, Mistral, NVIDI
TL;DR: GptOpen is an MIT-licensed AI chat app where you create custom AI agents (name, avatar, instructions, personality), chat solo or in group chats, and plug in ANY API model — Groq, OpenRouter, Gemini, Mistral, NVIDIA, DeepSeek, Kimi. No accounts, no database. Live: https://gptopen.vercel.app/ — source: https://github.com/ItsRayZer/axo-ai
Every "open" AI chat app I tried had the same fine print: sign up, hand over your data, chat inside their walled garden, pay their markup. So I built the thing I actually wanted — an AI chat app that's open in the ways that matter.
What GptOpen is
An agentic AI chat app. Not just a chatbot window — a place where you create AI agents, give each one a name, an avatar, instructions and a personality, and chat with them solo or together in group chats. Put three agents in one room and watch them discuss, react, and answer together.
Any API model. Your key, your rules.
This is the part I'm proudest of. GptOpen works with any API model:
- Groq, OpenRouter, Gemini, Mistral, NVIDIA, DeepSeek, Kimi
- Any OpenAI-compatible endpoint
Paste your own API key in Settings and chat unlimited — the only cost is your provider's usage. No markup, no middleman pricing, no "credits" system. And if you don't have a key, free built-in shared models work instantly, plus invite codes for unlimited shared use.
Privacy that's structural, not promised
- No accounts. No database. Your API keys and chats stay on your device.
- Keys are stored server-side via a Vercel relay — never baked into the page.
- The only thing that ever leaves your device is the API call itself.
The stack (devs will like this)
Vanilla HTML/CSS/JS. Zero build step. One serverless function (api/relay.js) handles the relay with automatic model failover. Fork it, connect to Vercel, add your API keys as environment variables, deploy. MIT licensed — build on it, self-host it, do whatever you want.
The UI is premium mobile-first: liquid glass, dark mode, smooth animations, Markdown rendering. I wanted it to feel like a flagship product, not a side-project demo.
Try it
👉 https://gptopen.vercel.app/
⭐ https://github.com/ItsRayZer/axo-ai
If you fork it or self-host it, I'd genuinely love to hear about it — drop a comment. And if you find a bug, the issues tab is open.
Disclosure: I'm the maker (Aben Sojan, 20, BCA student from Kerala, India). This is my project, MIT licensed, free.
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.