An uncensored AI chat with the limits written down
MadGPT is a chat with seven models and a router named Auto. It is for the question a mainstream assistant ducks: adult fiction, a disputed point of history, a blunt technical argument, security research on a system you are allowed to touch. You ask it. It answers.
Fewer refusals are the product. A blank check is not. MadGPT refuses help with crime. That covers sexual content involving minors, an attack on a system you do not have permission to test, fraud, and the rest of what is illegal where you are. The slot that answers more of the lawful questions does not move that line.
The picker shows seven models, plus Auto. Auto is a router, not an eighth model. It picks a route, and it does not select MadGPT Uncensored. The seven are MadGPT v4.0, v4.1, v4.3, Coder, v5.0, Uncensored, and v5.1. v4.0 is the fast slot. v4.1 and Coder are for code. v4.3 is the reasoning slot. v5.0 is an uncensored slot. MadGPT Uncensored is a separate uncensored reasoning slot. v5.1 is the flagship.
MadGPT Uncensored is on paid plans only. A trial account cannot select it, and a request for it is refused. It can read an attached image. It does not generate images. No model on MadGPT generates images. An image-generator query is a different product.
A filtered assistant often spends the reply explaining why it will not write the scene, the adult question, or the ugly side of an argument. MadGPT is pointed at those conversations. The answer is the model's answer. It can still be wrong. Medicine, law, and a patch you are about to ship need a check outside the chat. Taking the refusal layer off does not add knowledge the model never had.
Security research is part of the same stance. The work is penetration testing guidance, vulnerability analysis, and exploit development on a system you own or are allowed to test. A request to break into someone else's network is outside the product. Coder and v4.1 are the coding slots when the job is code. v5.1 is the flagship when you picked the hard problem on purpose. Auto routes ordinary chat. It does not route you onto MadGPT Uncensored.
The 1 Month, 3 Months, and 1 Year plans can upload 4 images a day and 4 text or code files a day. Lifetime: unlimited file and image uploads. Images can be jpg, jpeg, png, webp, or gif, up to 2 MB. MadGPT can read an attached image. It does not turn a prompt into a picture.
The chat itself stays at www.madgpt.io. This page is the explanation. The button goes there.
The plans are 1 Month at $99, 3 Months at $250, 1 Year at $999, and Lifetime at $2499.
Generations on those plans are unlimited.
Web search is included.
Credit packs are a separate purchase, and they do not include uploads.
Crypto at checkout is BTC, ETH, SOL, and USDT, plus a manual payment in Telegram.
Registration does not add credits.
The signup mail does not contain a key.
The Telegram bot issues a trial key for 7 credits.
The 7 days start at the first chat.
Paste the key on the site.
No credit card.
The trial cannot open MadGPT Uncensored.
Chats are stored on your account so the history is there when you return.
Deleting the account deletes that copy.
Chats are sent to third-party model providers to produce the reply, and those providers have their own retention policies.
This app does not run its own training job on them.