Never says no.
An open model with its refusal reflex removed. Ask it what other assistants won't touch.
The modelSealed.
It runs inside an encrypted machine. The cloud that hosts it can't read your conversation.
See the pathProven.
The machine signs a public report of what it is. You don't take our word for it, you check it.
How the proof works
Rent your own sealed GPU
A Sealed Box is a private machine with the uncensored model of your choice, billed by the hour. No message limit, nobody else on it, wiped when your time ends.
| Model | GPU memory | Price | Rent |
|---|---|---|---|
| Gemma 4 26B A4B · HereticLowest price | 48 GB GPU memory | $1.99 /hr−30% $1.39 with the token | Rent |
| Dolphin Mistral 24B · Fine-tuned | 48 GB GPU memory | $1.99 /hr−30% $1.39 with the token | Rent |
| Qwen 3.8 27B · HereticLive | 80 GB GPU memory | $2.99 /hr−30% $2.09 with the token | Rent |
| Gemma 4 31B · Heretic | 80 GB GPU memory | $2.99 /hr−30% $2.09 with the token | Rent |
Prices per hour in ETH at the live rate. Renting opens at launch. 6 boxes in the marketplace.
Sealed, end to end
Follow one question from your screen to the sealed machine and back. The host never holds a key, and the machine proves what it is.
You ask. It leaves encrypted.
Don't trust us. Check.
Every answer is computed inside a sealed machine: an NVIDIA GPU in confidential mode, behind Intel TDX. Your conversation is encrypted in use, so the machine's host can't read it. The attestation is public, and anyone can verify it.
How to check the proofRunning now
Everything about the modelQwen 3.8 27B · Aggressive Live
- Base
- Qwen 3.8, 27B, open weights
- Freed with
- Aggressive: the refusal direction is removed from the weights
- Runs on
- One sealed machine, NVIDIA GPU in confidential mode
- Free trial
- 5 messages, 600 characters each, no wallet
Stake it. Own a slice of the machine.
Stake
Every day we turn a share of what our sealed GPUs can serve into credits. 80% go to stakers, split by stake. No lock.
Earn credits
One credit is about 750 words from the uncensored model. Talk with them, or plug them into your tools with an API key.
Sell them for ETH
Don't need them? List them on the market. 40% of what the house earns buys the token back and burns it.
Milestones
- Now
Genesis
Qwen 3.8 27B, freed and sealed. 100 seats.
- $1M
Cluster
1,000 seats. A second sealed GPU.
- $5M
Vault
GLM 5.3, freed. Sealed Boxes on a dedicated GPU.
- $10M
Singularity
Uncensored coding and image models.
A milestone unlocks when the market cap holds for a week and fees can pay for the hardware.
What comes next
- Coming
Weights fingerprint
The attestation will also carry the hash of our exact model weights, so you can check which model answered.
- Coming
Hourly refusal test
A public benchmark of refusal-bait prompts, run every hour and signed by the enclave.
Questions
Is it really uncensored?
The model answers what mainstream assistants refuse: no lectures, no moralizing, no "I can't help with that". One line stays closed, and it's enforced by our code, not left to the model: nothing sexual involving minors.
Can you read my chats?
The model runs in a sealed machine, so the cloud that hosts it can't read them. Your messages pass through our server only to count your free messages on the way in. We don't store your conversations.
Do I need a wallet?
Not to try it. Five free messages, no sign-up, no wallet. You only connect a wallet to open your own Sealed Box, and the wallet opens only when you click.
What does the token do?
Stake it and you earn credits every day: a share of what our sealed GPUs can serve, split by stake. Use them to talk, sell them for ETH, or spend them through the API. 40% of the ETH from credit sales buys the token back and burns it, and Sealed Box hours paid with the token burn too.
What is Genesis?
The first 48 hours after launch. Tokens staked during Genesis earn 2× credits until it ends.
Why 18+?
Because an assistant that doesn't refuse will talk about adult subjects. The free chat is for adults only.