Models we run: DeepSeek R1 671B Llama 4 Maverick Mistral Large 3 Kimi K2.6 All open weights. We control the inference stack end-to-end.

Workspaces = persistent Kubernetes pods: - 2-8 vCPU, 2-32 GB RAM - Starts in <60 seconds - Auto-commits to private git - Access via browser, VS Code, SSH