Docs are being written alongside the model.

Norvaex is an AI research lab building Auxilium, a foundation model written from scratch in C. Norvaex is not a wrapper or a fine-tune of an existing model — the architecture, training stack, and runtime are its own.

Auxilium is the foundation model Norvaex is building. It is designed for deep, long-running work — multi-step research, large-codebase reasoning, and sustained analysis — rather than short conversational replies.

Writing the engine in C removes layers between the model and the hardware, which makes inference cost and latency predictable and keeps the runtime small enough to audit and self-host. Performance and privacy become properties of the architecture instead of configuration options.

Not yet. Auxilium is in pretraining. Norvaex publishes documentation and results when the interfaces they describe are stable, rather than announcing milestones in advance.

Auxilium is built to an enterprise standard of reliability but is not restricted to enterprise customers. Norvaex describes this as enterprise-grade, not enterprise-only: individual developers and researchers get the same model.

Norvaex treats customer data as belonging to the customer. Data is not used to train shared models without explicit opt-in, the model can be deployed in a customer's own environment, and Norvaex commits to never building for autonomous weapons or mass surveillance.