Hybrid LLM Gateway
Sensitive work on open-weight models in your own account, other work on the cloud's model service, all behind one gateway with budgets and audit.
One gateway in front of every model. It knows who is calling, checks prompts and answers, applies each team's token budget and logs every call. Work on sensitive data goes to open-weight models served on GPU Kubernetes inside the customer's account; other work goes to the cloud's managed model service in the same region. Where data must stay in Vietnam or Japan, an in-region endpoint from the FPT group is one more route. A reference design for regulated workloads.
Point at a component to light its links; select it for what it does here and the practice that makes it right on AWS.
When to use it
- Some data may never leave the customer's account, but other work needs frontier models
- Many teams and agents call models, and spend has to be visible and capped per team
- A model or provider must be swappable without changing the applications that call it
How NGT delivers it
Built from code, reviewed at a gate by an NGT architect, and handed over with runbooks your team can run.
AI & Agents
