Veylaro
Sign in Download
Licences & notices

What Veylaro is built on.

We think you should know exactly what's inside the thing running on your machine. Here's every piece, and the licence it arrives under.

The short version. Laro is built on Google's Gemma open-weight models. The current preview uses Gemma 3 GGUF quantizations, governed by Google's Gemma Terms of Use and Prohibited Use Policy, not Apache 2.0. Veylaro's agent and application software are separate from the upstream model weights.

1. The base model

The current download catalogue pins Gemma 3 4B, 12B and 27B instruction-tuned GGUF quantizations for Laro Lite, Med and Max. It does not establish a separately trained Veylaro adapter or base model. Using these weights carries upstream obligations:

  • The Gemma Terms of Use govern these Gemma 3 weights and applicable derivatives.
  • We pass along Google's Gemma Prohibited Use Policy to you as a user of a Gemma-derived model, exactly as the terms require.
  • Quantization and packaging do not make the upstream weights Veylaro-exclusive.
  • Google retains the ability to restrict uses that violate its policy.

Read them yourself: Gemma Terms of Use · Prohibited Use Policy · Google's Gemma 3 model card.

Future releases. Model families can have different licences. A future model replacement must include its own licence notices; it does not change the terms of weights already downloaded.

2. What is ours, and closed

Everything we added is Veylaro Labs' proprietary work and is not open source:

  • Veylaro-authored training infrastructure, where not covered by third-party licences.
  • The agent runtime: planning, verification, the adversarial ratchet, the existence gate.
  • The safety guard that stands between the model and your disk.
  • Veylaro Code (desktop + CLI) and this website.
  • The Veylaro and Laro names, the twin-blade mark, and the brand.

The upstream weights retain their own terms. Veylaro's application licence should not be read as replacing Google's model licence. Local learning is optional; no separate, qualified Veylaro fine-tune is claimed for this preview catalogue.

3. Your code and your data

You own everything you write and everything Laro writes for you. We claim no licence over your code, your prompts, or the output — none of it reaches us in the first place. See the privacy policy for the full detail.

4. Open-source components

Veylaro Code is an Electron application and ships with open-source software, each under its own licence — including llama.cpp/ggml, Electron, Node.js, React and Vite (MIT), and the Inter, Space Grotesk, Jost and JetBrains Mono typefaces (SIL Open Font License 1.1). The packaged THIRD-PARTY-LICENSES.md contains notices. A Settings shortcut to that file is not yet implemented in the current preview.

5. Benchmarks

Historical measurements and internal project checks are identified on the benchmarks page. They are not certified scores for the current downloadable models. Current-release comparisons remain pending until the model identity, task list, harness, budgets and execution results are published together.

6. Trademarks

Gemma and Google are trademarks of Google LLC. Veylaro Labs is not affiliated with, endorsed by, or sponsored by Google. "Veylaro" and "Laro" are trademarks of Veylaro Labs.

7. Questions

Think we've got something wrong on this page? We'd genuinely rather know. A support channel goes up alongside the public launch, and it'll be listed right here.

Last updated 1 October 2026.