We deploy locally by default when data can't leave.
Open-weight models (Llama, Qwen, Kimi, DeepSeek) running inside your VPC, on-prem or air-gapped. That means model selection and evaluation against your data, a deployment architecture, guardrails and monitoring, and a handover your own engineers can operate without us. When a hosted API is honestly the better answer, we say so and explain why.