ABOUT
DefiledAI is a community research hub for open-weight, abliterated, and uncensored local AI models. We build tools, publish benchmarks, and document the techniques that let people run powerful AI on their own hardware — without restrictions, without cloud dependency, and without per-query cost.
WHY THIS EXISTS
The local AI community has grown significantly in the past two years, but the resources available to it are scattered. Benchmarks live in Reddit threads. Model recommendations are buried in Discord servers. Beginner guides are outdated. Nobody had built a single place that serves both the person asking "what GPU do I need?" and the researcher experimenting with MoE pipelines.
There is also a specific gap around uncensored and abliterated models. These are legitimate research tools — the techniques involved (representation engineering, fine-tuning) are published academic work. The models are open-weight and legally distributed. The community using them deserves a proper home.
WHAT WE BUILD
OUR POSITION ON UNCENSORED MODELS
Open-weight models are software. Abliteration is a mathematical operation on weight matrices — the same class of technique as pruning, quantization, or fine-tuning. The models we document are legally distributed under open licenses. The techniques are published in peer-reviewed papers.
We document these models and techniques because the community using them deserves accurate information: quality benchmarks, hardware requirements, honest comparisons, and technical depth. That is the scope of what we do here.
We do not facilitate illegal activity. Content that crosses into illegal territory is not something we assist with regardless of the model or technique involved.
CONTRIBUTE
The site improves with community contributions. If you have benchmarks to share, models we missed, or tutorials to write — the forum and submission tools are the right places.