TRX22 · Platforms and Ventures
We architect high-efficiency AI platforms.
All of your organization's inference passes first through a layer that protects sensitive data, trims the request before the call, and records who did what, how much it cost, and what was blocked.
Security and governance applied at the point of passage, not in a report afterward.
Protection
before it's sent
Membrane always runs, with no switch to turn it off
Kernel
Rust + Python
TRX Core v2, compiled into the hot path
Telemetry
1 record / request
Cost, energy, blocks, and who's responsible
Partnerships
Claude Certified Partner
Inception Program
Who we are
A product company, not a consultancy advising from the outside.
The accelerated adoption of generative AI opened a structural gap: organizations lack the governance, security, and efficiency infrastructure between them and the models they use.
Proprietary technology validated by reproducible benchmark, plus a consulting arm that applies it in agile transformation and governance rollout.
Mission
Make AI adoption in organizations auditable, secure, and efficient — giving companies and governments the control they currently lack over how, where, and at what cost AI is used.
Vision
To be the reference AI governance infrastructure for the enterprise and government markets: the layer between any organization and any AI model it uses.
The platform
Three layers, built in this order.
The control plane first, because it creates value without requiring your own GPUs. National-grade execution next. The model last. Each layer only starts once the previous one is proven.
Control plane
TRX Canopy
The efficiency and governance kernel between you and AI. OpenAI-compatible gateway, deterministic DLP before dispatch, FinOps by area, and a sealed audit trail.
See the productExecution
TRX Sovereign Engine
Open-weight model inference on NVIDIA infrastructure hosted by partners in Brazil, with data residency and tenant isolation.
See the productModel
TRX Canopy.AI
A sovereign base language model for corporate workflows, code generation, and tool use by autonomous agents.
See the roadmapThe hard questions
What they ask before signing.
Short answers, including the ones that don't favor the sale. A verifiable "no" is worth more at a boardroom table than a "yes" that doesn't survive an audit.
Will my team have to switch tools?
No. The gateway is OpenAI-compatible, so what changes is the base URL — code you've already written keeps working. What the team gains is a virtual key per area, with its own budget, rate limit, and allowed models.
Does sensitive data leave my infrastructure?
Membrane inspects inside your network, before dispatch, and what it stores from a detection is a content hash. For those who can't let data leave the premises, there's Canopy On-Premise or Offline.
Does this make me compliant with LGPD and the AI Act?
No, and be wary of anyone who says it does. Canopy supports the requirements: it logs decisions, applies notices deterministically, and keeps the evidence. Certification and compliance guarantees aren't ours to give.
Who can delete the audit trail?
No one, not even administrators. The log is append-only and chained. And the panel also shows the calls where a control didn't apply, with the reason — because that's the list an audit asks for first.
Contact
Looking for our first Design Partners.
If your organization already runs AI in production and can't precisely answer how much it costs, what it processed, and who triggered it — that's exactly the conversation.
Companies
Mid-size and large organizations already running AI in production that need to answer how much it costs, what it saw, and who used it.
Government
Public bodies, with a direct procurement path under Brazilian Law 14.133 via CPSI, and an On-Premise option for data that can't leave the premises.