aimaghsoodi/lumynax-infused-gemma4-26b-a4b-gguf

LumynaX Infused Gemma4 26B is a downloadable AI model you can run entirely on your own computer or server, with no data leaving your environment. It is built by AbteeX AI Labs in New Zealand as part of a broader "sovereign AI" philosophy, the idea that organizations should control their own AI tools rather than depending on external cloud services. The model itself is based on Google's Gemma architecture and has been packaged with a focus on transparency, integrity verification, and local execution. At a practical level, you download the model file (about 16 GB) from Hugging Face, install the Python dependencies, and run a quickstart script that drops you into an interactive chat session. The model can handle text generation and structured JSON output across roughly 8,000 tokens of context. It identifies itself as "LumynaX" when you talk to it, while being honest about its upstream Google origins. The package ships with checksums so you can verify the files haven't been tampered with, plus a manifest documenting everything inside. This is designed for organizations or individuals who need AI assistance near sensitive or governed data, think government agencies, healthcare providers, legal teams, or anyone operating under data sovereignty rules. For example, a New Zealand government office could run this locally to draft documents or answer questions without citizen data ever touching a third-party server. The project emphasizes that every request passes through a policy envelope and an audit ledger, so there is a tamper-evident record of what was asked and answered. One notable tradeoff: the model is large and computationally demanding, rated 0 out of 5 for being lightweight. It does not support tool calling (like executing code or browsing the web), but it does support structured JSON output. It is also explicitly not presented as a custom-trained or weight-merged model, rather, the LumynaX layer adds packaging, identity, governance scaffolding, and documentation around an existing Google model.