Heretic: Uncensoring AI Models Made Simple, Open-source and Free.

amy 25/03/2026

What is Heretic?

Heretic is a straightforward tool that takes a standard AI language model and gently removes its built-in “I can’t help with that” reflex, without needing to retrain it from scratch or have a PhD in machine learning. Think of it like adjusting the sensitivity on a microphone: instead of muting certain topics entirely, Heretic fine-tunes the model’s internal settings so it responds more openly, while keeping its original knowledge and reasoning intact.

It works automatically: you point it at a model, and it searches for the right balance between being helpful and staying true to the model’s core abilities.

The result is a version that answers more freely, with less distortion than many manual tweaks. It runs on your own machine, supports popular models, and even offers options to save, test, or share your results.

No cloud required, no complex setup, just a command, a little patience, and a model that behaves more like a candid conversation partner.

How Does Heretic Work?

Imagine an AI model has an internal “brake pedal” it hits whenever a question feels off-limits. Heretic doesn’t rip out the brakes, it just adjusts how sensitive they are. It does this by first studying how the model reacts to sensitive questions versus everyday ones, spotting the subtle mathematical patterns behind its “I can’t help with that” reflex.

Then, layer by layer, it gently nudges those patterns aside, more in some places, less in others, so the model stays sharp and helpful without defaulting to refusal.What makes Heretic clever is that it doesn’t apply a one-size-fits-all fix.

It fine-tunes where and how much to adjust, even blending different “refusal patterns” together to find a sweeter spot. It also treats different parts of the model with care, knowing that tweaking some areas can hurt performance more than others, so it prioritizes changes that keep the model’s intelligence intact.

The whole process runs automatically. You don’t need to understand vectors, matrices, or transformer internals. Just pick a model, run the tool, and let it find the right balance between openness and ability, like a skilled mechanic tuning an engine by ear, not by blueprint.

License

Copyright © 2025-2026 Philipp Emanuel Weidmann ([email protected]) + contributors

This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more details.

You should have received a copy of the GNU Affero General Public License along with this program. If not, see https://www.gnu.org/licenses/.

By contributing to this project, you agree to release your contributions under the same license.

Resources & Downloads

  1. Source-code and downloads