abli.cc

Uncensored AI models, hosted for you.

abli.cc is building an easy-to-use, secure service for running abliterated — uncensored — open language models. No refusals, no lectures: just capable models that answer your questions.

Launching soon. Leave your email and we’ll invite you when the doors open.

An open birdcage releasing a glowing paper crane — an AI model set free
Out of the cage: abliteration removes the refusal reflex, nothing else.

What is an “abliterated” model?

Plain-language summary of Maxime Labonne’s article “Uncensor any LLM with abliteration” on Hugging Face.

Modern AI assistants are trained twice: first to be knowledgeable, then to refuse certain requests — the familiar “As an AI assistant, I cannot help with that.” That refusal behavior is bolted on top of the model’s real abilities.

Researchers discovered something surprising: inside the model, the entire urge to refuse lives in one single “direction” — like one fader on a mixing desk. Turn that fader down and the model simply stops refusing, while keeping all of its knowledge and skills intact.

Abliteration is the surgical technique that finds that fader and removes it — no retraining required. The result is an abliterated (uncensored) model: the same brain, minus the built-in “no”. Read the full technical walkthrough on the Hugging Face blog.

In short

Is it a different model?
No — it’s a well-known open model with one internal switch flipped.
Does it lose intelligence?
No. Abliteration targets only the refusal mechanism; capabilities stay the same.
Why would I want one?
Because censored models often refuse legitimate work — see below.

What people use them for

Censored models decline plenty of perfectly legitimate requests. Abliterated models don’t — and abli.cc will make them trivial to run.

  1. Test your site’s security

    Probe your own websites and apps for vulnerabilities with an assistant that won’t clam up mid-audit. Censored models routinely refuse security work; ours will help you find the holes before attackers do.

  2. Create without a filter

    Write free-speech text, generate images and produce video with models that treat you like an adult. Fiction, satire, art, research — your prompt, your call.

  3. And much more

    Debate prep, red-team exercises, edgy marketing copy, honest second opinions, research into model behavior itself. If a mainstream chatbot has ever told you “I can’t help with that” — this is the alternative.

Explore the models

Abliterated models are open and public. Browse the community collections on Hugging Face to see what’s out there — these are the kinds of models abli.cc will host for you.

#abliterated Models surgically uncensored with the abliteration technique — sorted by trending. #uncensored The broader uncensored collection: abliterated models plus other unfiltered fine-tunes.

New to this? Start with the trending tab on either page: download counts and likes are a decent proxy for quality. Names mentioning Llama, Qwen or Mistral are built on well-known open foundations.

Going deeper? Labonne’s article links FailSpy’s abliterator library and a curated collection of abliterated models — or just wait for abli.cc and skip the GPU plumbing entirely.

Be first through the door.

We’re setting up secure, hassle-free hosting for uncensored models right now. Register your interest today and you’ll be among the first invited — early supporters will get founding-user pricing for life.

Prefer email? Write to hello@abli.cc with subject “Keep me posted”.