TRX Sovereign Engine
Sovereign inference, run in Brazil.
TRX Sovereign Engine is an AI inference platform in development for regulated companies and government. It will run open-weight language models on NVIDIA infrastructure hosted by partners in Brazil, offering data residency, tenant-isolated service, predictable capacity, and integration with TRX Canopy's governance, DLP, FinOps, and audit controls.
The first release focuses on shared and reserved inference for 8B–70B models, with larger MoE-architecture workloads added after benchmark validation.
Residency
Brazil
NVIDIA infrastructure hosted by national partners
Isolation
tenant-isolated
Service isolated per tenant, no cache crossing boundaries
Models
8B – 70B
Open-weight, shared or reserved; MoE after benchmarking
Under the same control plane
The Engine isn't a second platform to govern. It answers to the same controls as Canopy: the policy that applies to the external provider applies to national capacity too.
Governance
The same Organization, Area, Project, and virtual Key hierarchy.
DLP
Membrane inspects before dispatch, including for the Engine.
FinOps
Cost by area and by owner, with predictable capacity.
Audit
The same append-only, chained trail, no exception by origin.
Caveat
Sovereign Engine is in development. Capacity, latency, and throughput per workload will be published with their own measurement — no performance figure before the benchmark.