The same agent platform powering trdn.io — deployed inside your own infrastructure, your own region, your own GPUs. No external API calls. No data egress. Built for the organizations that legally cannot route data offshore.
When the data is citizen records, account holders, or patient histories, "send it to OpenAI" isn't an option — it's a compliance violation. TRD Sovereign runs the entire agent stack where the data already lives. The engagement model: $5k–50k/month per deployment, scoped per environment.
Deploy in your national cloud. Citizen data stays inside the country. Air-gap-friendly — the pod runs with zero outbound dependencies beyond an optional license check. UAE government agencies are a core target.
Meet RBI, MAS, DFSA, and EU regulator requirements. Run AI inference on data you legally cannot expose to a third-party endpoint. Indian PSU banks are a primary deployment target.
HIPAA-compatible deployment. Patient records never leave your private network — inference happens inside your perimeter, not someone else's. EU healthcare systems are an active focus.
No surprise integration work, no open-ended consulting. The path from first conversation to a deployed Sovereign pod is three defined stages — and the model swap, isolation mode, and region are all your choices.
A 30-minute scoping call covers your environment, hardware, and compliance needs. If it's a fit, a paid pre-deployment infrastructure audit follows — and the fee converts to a deposit on a signed contract.
30-min call → paid audit
You fill a single config file — your domain, your model choice, your tenant isolation mode, your region. The deployment package handles the rest: backend, inference, database, monitoring, runbooks.
edit config.yaml → helm install
The full stack comes up in your Kubernetes cluster or via Docker Compose. The only outbound connection is an optional license callback — it carries a token, never data. You hold the keys.
trd-sovereign status → all green
The marketing surface, the lead pipeline, the architecture decisions, and the own-GPU inference layer are all built. Everything below is either live in production or actively in flight — tagged honestly.
The public-facing surface explaining the self-hosted proposition to enterprise prospects — backend lead-capture API, frontend content, protected-zone enforcement. Shipped May 9.
The sovereign_leads backend captures and qualifies inbound enterprise leads — validating company size, use case, and budget signals, with high-value leads routed for fast follow-up.
Six non-negotiable architecture decisions identified for any enterprise deployment — tenant isolation mode, encryption envelope at rest, append-only audit log, and the inference-router abstraction among them.
vLLM serving Qwen3-32B-FP8 — benchmarked at 1,605 tokens/second aggregate at concurrency 16. This is the layer that makes "sovereign" real: inference on hardware you control, not someone else's API.
A Qwen3-0.6B draft model proposes tokens for the 32B model to verify — 66% acceptance rate, 25%+ throughput gain. Production-grade serving, not a research demo.
The own-GPU inference layer that becomes the primary engine for Sovereign deployments. Code shipped; currently stabilizing on dedicated H200 capacity before it's the default path.
Sovereign is deliberately sequenced — the engineering for compliance is done up front, certifications are pursued only when a paying customer requires them, and multi-region automation comes once the deployment package is proven. Honest phasing, not vapor.
A focused design week to lock the six architecture decisions before any enterprise deployment is offered: tenant isolation, the encryption envelope and per-customer key management, the append-only audit log, and the inference-router abstraction with backwards compatibility.
The one-package install for enterprise customers. Docker Compose for single-server deployments, a Kubernetes Helm chart for larger ones — bundling the TRD backend, vLLM with the chosen model, self-hosted database, storage backup, monitoring stack, and ops runbooks.
For enterprises that want premium model quality but can't go fully on-prem: an Anthropic Zero Data Retention API tier, where request data is never stored. A middle path between full sovereignty and standard cloud APIs.
The first commercial Sovereign customer — target: a UAE government agency, Indian PSU bank, or EU healthcare system. Engagement model: a 30-day proof of concept, then an annual contract with deployment, training, SLA support, and quarterly reviews.
Formal certifications — pursued only after paying customers explicitly require them. The underlying engineering (audit logs, access controls, encryption) is done up front in the Phase 2 locks, so certification becomes audit prep, not a rebuild.
Scale-out for serving multiple Sovereign customers across regions. A customer signs up, picks a region, and the system provisions a pod, configures their tenant, and returns connection details — UAE, EU, India, and US targeted.
Cloud AI APIs are fast to start with. They also send your data to someone else's servers, in someone else's jurisdiction, under someone else's terms. For regulated organizations, that's not a latency tradeoff — it's a non-starter.
| Property | TRD Sovereign | OpenAI / Anthropic API | Cloud-hosted LLM (Bedrock / Vertex) |
|---|---|---|---|
| Data stays inside your perimeter | ✓ | × | ~ |
| Runs in your region / national cloud | ✓ | × | ~ |
| Air-gap-capable deployment | ✓ | × | × |
| Zero external API calls at inference | ✓ | × | × |
| You hold the model weights & keys | ✓ | × | × |
| Full agent stack — 641 agents, not just an endpoint | ✓ | × | × |
| Observability exportable to your SIEM | ✓ | × | ~ |
| No usage data retained by a vendor | ✓ | ~ | ~ |
| Predictable cost — flat monthly, no per-token metering | ✓ | × | × |
| Standard tooling (Helm / Compose) | ✓ | × | ✓ |
"~" denotes partial: some cloud-hosted LLM offerings keep data in-region but still run on vendor-controlled infrastructure under vendor terms. Sovereign removes the vendor from the data path entirely.
Sovereign isn't a per-seat SaaS — it's an infrastructure deployment. Every engagement starts with a paid audit, and the audit fee converts to a deposit on a signed contract. The figures below are the engagement structure; exact scope is finalized per environment.
A pre-deployment audit of your cluster, hardware, network topology, and compliance requirements. Not a sunk cost — the fee converts to a deposit on a signed contract.
A full Sovereign pod deployed in your environment for a 30-day proof of concept — the real stack, your data, your perimeter, before any annual commitment.
An annual contract per deployment — $60k–500k ARR depending on scale, model, and support tier. Includes the deployment, training, SLA support, and quarterly business reviews.
All figures reflect the engagement structure; exact scope and pricing are finalized in the infrastructure audit. The first deployments are founder-led.
We pursue formal certification when a specific deployment requires it — the engineering work (audit logs, access controls, encryption) is done up front in the Phase 2 architecture locks, so certification is audit prep, not a rebuild. We're happy to share the gap analysis for your jurisdiction first.
Need a framework not listed here? The pod's architecture — no egress, your keys, your infrastructure, your region — is designed to map cleanly onto most data-residency and sovereignty regimes. Ask us for the gap analysis for your jurisdiction.
The first step is a 30-minute call to scope your deployment. If it's a fit, we run a paid pre-deployment infrastructure audit — which converts to a deposit on a signed contract.