Is it really private?
Yes. All inference runs on open-weight models (Qwen, Llama, Mistral, DeepSeek) on our own server. Your prompts never go to OpenAI, Google, or Anthropic. We don't keep logs, and we never train any model on your data.
Which models do I get?
Qwen 3 (4B and 14B), Llama 3.1 8B, Mistral 7B, DeepSeek-R1 8B, Gemma 2, and more. Pick per-conversation. If a larger model is slow to start, we auto-fall back to a fast local model — never to the cloud.
How does the 50% referral credit work?
Your username is your referral code. Share rochatgpt.com/?ref=yourname. When a friend subscribes through your link, every monthly invoice they pay adds 50% (10 MXN) to your Stripe customer credit balance. That credit is automatically deducted from your next bill. Refer 2 active friends and your own subscription is effectively free.
Can I cancel anytime?
Yes. One click in the Stripe customer portal. No calls, no retention tricks. You keep access until the end of your paid period.
What if the model is slow?
If the large model takes too long to start responding, we automatically route your request to a faster local model. You'll see a small notice when that happens — and we never fall back to any cloud provider.
Do you have a free trial?
Not today — running local LLMs 24/7 costs real money. But 2 referrals covers your whole subscription, so the first month can be your only paid month if you invite a couple of friends.