Client
Privacy-sensitive organization
Service
Implement
Status
Shipped
Stack
Local open-weight models · sensitivity router · eval harness
A privacy-sensitive org's on‑prem AI.
An AI assistant and agent system on the client's hardware. A sensitivity-based policy controls which tasks run locally and which may use cloud models.
The brief
A privacy-sensitive organization wanted AI assistance while keeping restricted data inside its own network. It also wanted the option to use external models for tasks permitted by its data policy.
What we built
We built a routing layer that determines where each request runs according to the client’s sensitivity policy. Restricted workloads use open-weight models on the client’s hardware. Permitted tasks can use external models. The client controls the policy, and requests are logged for review.
On-prem · your network
Your team
Routerby sensitivity
Local modelssensitive data stays here
non-sensitive · opt-in
External cloud APIs
Testing and data controls
A repeatable set of tests checks local model performance against the client’s requirements and supports evaluation when models change. We sized the hardware for the workload and used compressed model formats suited to those machines.
Delivery
We delivered the system with documentation so the client’s team could operate it and adjust the routing policy. Logs provide a record of routing decisions, while local and external models serve the tasks permitted by that policy.
Sensitive dataLocalstays on your hardware
What leavesYour callpolicy-based routing
Every requestLoggedfull audit trail
