Tartarus
Sovereign Edge
Tartarus is the security-hardened edge sibling to the sovereign local-AI platform. A tri-silicon rig — CPU, integrated GPU, discrete GPU, and a dedicated NPU for roughly 300 TOPS of combined throughput — runs inference with zero cloud cost, backed by a local object store on the T7 Shield. It dispatches across CUDA, XPU, vLLM, and OVMS runtimes behind post-quantum-secured tunnel ingress and TPM hardware attestation, forming a two-tier sovereign stack: device plus hardened edge. An 8-squad agent matrix and zero-egress loopback keep serving traffic inside the sovereign boundary.
What it does
Multi-Silicon Dispatch
Intelligent routing of inference across CUDA, XPU, vLLM, and OVMS runtimes based on workload shape and availability.
Post-Quantum Tunnel
Ingress terminates post-quantum-secured tunnels, defending transit against store-now-decrypt-later adversaries.
TPM Attestation
Every worker proves its boot state to a hardware root of trust before it is admitted into the serving pool.
Zero-Egress Loopback
Serving traffic never leaves the sovereign boundary — internal hops traverse private loopback, not public networks.
- Tri-silicon (CPU + iGPU + dGPU + NPU)
- Zero cloud cost · local object store
- Multi-silicon dispatch
- Post-quantum tunnel ingress
- TPM hardware attestation
- Post-quantum tunnel ingress
- TPM hardware attestation
- Zero-egress loopback
- Two-tier sovereign stack