Local AI inside the engineering boundary

Public cloud coding assistants are convenient, but controlled engineering programs have a different first principle: source code and technical data must stay inside the approved environment.

Tetryx is designed around that boundary. The product is positioned as a containerized software appliance for local GPU servers, with open-weights models serving IDE and terminal workflows without sending code to a public model API.

The boundary is the product

For regulated teams, model quality is only useful after the deployment model is acceptable. The practical questions arrive early:

• Where does code travel?
• Which systems can reach the model endpoint?
• What request data is retained for audit?
• Which IDEs and agents can use the local service?

The blog and docs in this repo now use local MDX for the same reason. The content is in the repository, versioned with the product surface, and no longer depends on a Notion export path.

What should be documented next

The next layer of documentation should come from real installation runs, real policy decisions, and real deployment diagrams. If a command, metric, or workflow cannot be reproduced by the team, it should stay out of production copy until it can.