Licenses
Nearly every OpenCerebral model, including its weights, tokenizer, and configuration, is released under the Apache License, Version 2.0. Copyright 2026 Joseph Jones. The single exception is Lucy-2, which is research-only because its training data is — see below.
You may obtain a copy of the license at apache.org/licenses/LICENSE-2.0.
| Artifact | License |
|---|---|
| Boris weights, all sizes and variants | Apache 2.0 |
| Boris GGUF builds | Apache 2.0 |
| littlerock-1M and littlerock-1M-arithmax | Apache 2.0 |
| Lucy | Apache 2.0 |
| Lucy-2 | Research use only |
| This website | Verbatim copying permitted, see footer |
What that means in practice
- Commercial use is permitted. There is no research-only clause.
- Modification, finetuning, quantization, and redistribution are permitted, including of derivatives you do not open-source.
- There is no acceptable-use policy attached and no field-of-use restriction. We consider licenses that carry those to be non-free.
- You must preserve the copyright notice and license text, and state significant changes.
- The software is provided on an "AS IS" basis, without warranties or conditions of any kind.
Lucy-2 and CelebA
Lucy-2 was trained partly on CelebA, which is licensed for non-commercial research use only. That term follows the model, so Lucy-2 is released for research use only and must not be used commercially. We would rather state that than pass a permissive license over data that does not carry one. The unconditional Lucy model is Apache 2.0.
Training data
The models are trained on FineWeb-Edu and DCLM-baseline, with smol-smoltalk, OASST1, and (for Boris 1.0 instruct models) Alpaca used for instruction tuning. Those datasets carry their own licenses and terms, published by their respective maintainers; our Apache 2.0 grant covers the weights and code we produce, not the upstream corpora.
Parent project
The Klondike Software Project licenses its other software under the Kunix License and the Klondike Public License family. OpenCerebral uses Apache 2.0 instead because it is the established convention for model weights and the most broadly compatible permissive choice.