EU AI Act ready · ISO 42001 in audit · accepting infrastructure audits

The Agentic Trust Layer
for AI that never leaves your perimeter.

The same agent platform powering trdn.io — deployed inside your own infrastructure, your own region, your own GPUs. Every inference, every action ed25519-signed and chained on a public provenance ledger. No external API. No data egress. Built for organizations that legally cannot route data offshore and need to prove what their AI did.

0
external API calls
641
agents, fully on-prem
1,605
tok/s · own-GPU benchmark
100%
receipt-anchored actions
EU AI ACT · READY
5 REGIONS
🛡
SOVEREIGN POD
$5–50k
/ mo · ARR
🇺🇸
US
HIPAA
🇪🇺
EU
GDPR
🇦🇪
UAE
Khazna
🇮🇳
India
DPDP
🇦🇺
ANZ
Privacy
Runs on the open inference stack — inside your environment
vLLMQwen3-32B-FP8Kubernetes 1.28+ Docker ComposePostgres + pgvectorPrometheus + Grafana
F13 · Sovereign Marketplace · preview

Curated models, one-click sovereign deploy.

Pick a model from the catalog, click deploy, get a sovereign endpoint in your VPC within ~6 minutes. Each model card includes the licensing tier, hardware requirements, and the receipt-chain attestation showing the exact build that was deployed. Catalog seeded with frontier open-weights — Llama family, Qwen, Mistral, DeepSeek, plus TRD-tuned variants.

All models · 10 Frontier · 4 Mid-tier · 4 Specialised · 2
Frontier

Llama 3.3 · 70B Instruct

● Available
Meta. Open weights. Best general-purpose 70B available. License: Llama 3 Community.
2× H100 ~6 min deploy
Frontier

Qwen 2.5 · 72B

● Available
Alibaba. Open weights. Multilingual leader (29 langs). License: Tongyi Qianwen.
2× H100 ~6 min deploy
Frontier

Mistral Large 2

● Available
Mistral AI. 123B params, strong code + EU jurisdictional comfort. License: MRL.
4× H100 ~8 min deploy
Frontier

DeepSeek V3

● Available
DeepSeek. MoE 671B (37B active). Best cost-per-token in the catalog. License: MIT.
8× H100 ~12 min deploy
Mid-tier

Llama 3.3 · 8B Instruct

● Available
Meta. Good cost-per-token for routine tasks; deploy on a single GPU.
1× L40S ~3 min deploy
Mid-tier

Qwen 2.5 · 14B

● Available
Multilingual mid-tier. Excellent for India / China / EU regulated deployments.
1× A100 ~4 min deploy
Mid-tier

Mistral Small 3

● Available
24B. Apache 2.0 license — unrestricted commercial use. EU-friendly default.
1× A100 ~4 min deploy
Mid-tier

Phi-4

⏳ Beta
Microsoft. 14B reasoning-tuned, runs on consumer GPUs. License: MIT.
1× L40S ~3 min deploy
Specialised

TRD-tuned · Clinic NLP

⏳ Coming
Llama-3 8B fine-tuned on clinical intake. HIPAA-aware deployment template.
1× L40S HIPAA preset
Specialised

TRD-tuned · Legal EU

⏳ Coming
Mistral Small 3 fine-tuned on EU contract law. GDPR + EU AI Act compliant.
1× A100 EU jurisdiction

Phase 1: catalog visible + deploy-flow scaffolded. Phase 2: one-click sovereign deploy backed by our pod federation infra (target Q3 2026). Want a model not listed? Email sovereign@trdn.io and we'll evaluate adding it.

Sovereign Data Vault · live today

Customer data, cryptographically provable end-to-end.

Sovereign is not a marketing label — every byte of customer data your tenants give us is wrapped, signed, and pinned in a way that makes the word "sovereign" mathematically true, not a vendor promise. Here is what's underneath.

Layer 1 · Encryption at rest

AES-256-GCM per-tenant DEK

Every tenant gets a unique Data Encryption Key. DEKs are wrapped by a tenant-specific Key Encryption Key (KEK). Shredding the KEK shreds all data for that tenant — cryptographic erasure, GDPR-clean by construction.

Layer 2 · Permanence

Filecoin pinning · same chain

Pinned via Lighthouse SDK to Filecoin. CIDs are content-addressed. Three SLA tiers — Bronze (1×, 1y), Silver (3×, 7y), Gold (5×, 99y). Replica health surfaced live in the permanence meter.

Layer 3 · Provenance

Ed25519 signed receipt chain

Every read, write, erasure, model call is signed at emit time and chained with prev_receipt_cid. Anyone can recompute the bytes and verify against the public key at trdn.io/keys/trd-receipt-key-2026.public.pem.

Layer 4 · PII isolation

PII vault · never crosses pod boundary

PII columns identified at write time, encrypted with the per-tenant DEK, and never leave the pod. Inference is performed against tokenized values; only the operator's own pod can decrypt.

Layer 5 · Cross-product permanence

Build artifacts auto-pinned

Compute build logs, agent outputs, image bank entries auto-pinned to the same chain. Prompt → compute → SITE → artifacts — everything content-addressed on Filecoin. Federation-ready.

Layer 6 · Public verification

verify.trdn.io · client-side

Customers, regulators, auditors paste a session ID. Browser validates every signature client-side against the published public key. Zero trust required in TRD servers. The trust is in math.

Six layers, every layer auditable, every action signed. The Sovereign Data Vault is what makes "Sovereign" technically provable instead of a vendor promise. Pair it with Compliance-as-a-Service for the audit-prep package.

New offering · F14 · Day 12 v3 Day 3

Compliance, productized.

The audit-prep work we already do for Sovereign customers — SOC 2, ISO 27001, ISO 42001, HIPAA, EU AI Act, GDPR — now available as a standalone engagement. Buy the package, get a 90-day audit-ready posture without spinning up a Big-Four contract.

Tier 1 · Readiness

Audit Readiness Pack

$4,900
one-time engagement
  • SOC 2 Type I scoping doc + control mapping
  • HIPAA technical safeguards review (if applicable)
  • GDPR Art. 17 / 20 erasure proofs via signed receipts
  • Asset inventory + data flow diagram
  • One follow-up Q&A session
Request engagement →
MOST PICKED
Tier 2 · Audit-Ready

90-Day Audit-Ready Posture

$14,900
one-time · 90-day delivery
  • Everything in Readiness Pack, plus:
  • ISO 27001 + ISO 42001 (AI Mgmt) gap assessment
  • EU AI Act risk classification + documentation
  • Policy templates: Acceptable Use, Data Retention, Incident Response, Model Lifecycle
  • Vendor / sub-processor register
  • 2 audit dry-run sessions with auditor-style questions
  • Sovereign deployment hardening checklist
Request engagement →
Tier 3 · Ongoing

Continuous Compliance Retainer

$1,900/mo
12-month minimum
  • Quarterly control re-review
  • Receipt chain audit report (auto-generated monthly)
  • Vendor risk re-assessment
  • Policy refresh as standards evolve (EU AI Act, ISO 42001 amendments)
  • Incident-response standby (1-business-day SLA)
  • Annual mock-audit walkthrough
Request engagement →

Why this works for TRD Network

Every action a TRD-Sovereign customer takes is already ed25519-signed and pinned to Filecoin. The receipt chain is the audit log — we don't have to manufacture evidence retroactively because it exists at emit time. That cuts audit-prep work for a 50-control SOC 2 from 6 weeks (typical) to 7 days (TRD baseline). The price difference between "Big Four readiness" ($60k–$120k) and the Audit-Ready Posture above ($14,900) is the moat made concrete.

Engagement requests routed to compliance@trdn.io. We respond within one business day with a scoping call invitation. Stripe checkout link goes live once SKU is provisioned — request a manual invoice in the meantime.

Day 12 v3 · Sovereign · shipped this week

What's live now, in-flight, and next.

● Live FEDERATION

Pod-to-pod attestations

Federated chain-head exchange. Sovereign pods cross-attest each other's receipts — proof of cooperation without trust.

● Live INVITE TOKENS

HMAC-signed pod invites

JWT-style tokens with HMAC-SHA256 signing. Operators issue scoped invites without revealing federation secrets.

● Live x402

x402 paid dispatch

USDC/USDFC/FIL dual-rail payment on the dispatch endpoint. Sovereign customers pay-as-they-compute, no SaaS contract required.

● Live PII VAULT

Per-tenant PII vault

PII never crosses the pod boundary. Per-tenant DEKs + cryptographic erasure built in — GDPR-clean by construction.

◐ In flight GOVERNANCE PANE

Operator governance UI

Quorum voting on federation membership + policy change proposals. Foundation laid; UI surface lands next sprint.

○ Next REGULATED VERTICALS

Healthcare + finance kits

Pre-built sovereign deployment kits with HIPAA / SOC2 / GDPR audit trails. Receipt chain doubles as the audit log.

● Live F13 · MARKETPLACE

Sovereign Marketplace · real deploy lifecycle

Pending → provisioning → ready with auto-advance cron every 5 minutes. Operators see real deploy state on the model catalog, not a placeholder.

● Live F1 · REPUTATION SBT

Reputation SBT · Base mainnet

Contract 0xbDd1f5fC349D9a8EfCEb07Edbd491233b2540f5F live on Base. First mint executed. Sovereign-operator reputation now anchored on-chain alongside the receipt chain.

● Live EU REGION

EU passive region · api-eu.trdn.io

trd-backend-eu deployed. GHCR images public: ghcr.io/trdnetwork/{orchestrator,receipt-signer,ipfs-gateway,predictive-engine}:0.1.0 — sovereign customers can pin to EU data sovereignty.

00 — The Trust Substrate

Four cryptographic moats. One platform.

Most AI vendors say "we're compliant." TRD Sovereign proves it on every request. The platform ships with four production-grade trust mechanisms, each independently verifiable.

01 · PROVENANCE CHAIN

Signed Receipts on Every Inference

Every GPU call, every site build, every owner-command emits an ed25519-signed receipt. Receipts hash-chain forward and anchor to Filecoin. Tampering is mathematically detectable. Verify any chain at api.storage.trdn.io/api/storage/receipts/<id>.

02 · CRYPTOGRAPHIC ERASURE

GDPR Article 17 in One Click

Per-tenant DEK encrypts all PII at rest. Destruction zeros the DEK — every encrypted blob becomes mathematically unrecoverable. Customer downloads a PDF Certificate of Erasure. Satisfies GDPR Article 17, CCPA §1798.105, and regulator subpoenas in under 30 seconds.

03 · CARBON ATTESTATION

Energy Footprint, Per Request

Every inference receipt carries kWh + CO₂ grams calculated from GPU class × duration. Aggregated on cert PDFs. Live carbon badge on every built site. Required reading for any organization with a public sustainability commitment.

04 · C2PA CONTENT CREDENTIALS

Browser-Visible Verification

Same standard Adobe, Microsoft, BBC, and Reuters use. Every output page embeds a JSON-LD provenance manifest. Modern browsers and verify.contentauthenticity.org surface a "verified content" badge. Disinformation resistance, built in.

Full posture documented in the EU AI Act + ISO/IEC 42001 Readiness Statement

Read the readiness statement →
08 — Bring Your Own Model

Your weights. Your perimeter. Our trust substrate.

Sovereign customers run real-world inference on real-world models. Some have fine-tuned Llama 3 on a decade of internal medical records. Some signed multi-year licenses with Cohere or Mistral. Some have weights too sensitive to ever leave the pod. TRD Sovereign hosts all three, with the same signed-receipt + cryptographic-erasure substrate that makes Sovereign Sovereign.

BYOM · BRING YOUR OWN MODEL

Your fine-tuned weights, our runtime.

Bring an HF/PyTorch checkpoint, a GGUF quantized binary, or a TensorRT-compiled engine. TRD's vLLM + TGI runners ingest it, expose an OpenAI-compatible inference endpoint, and emit signed receipts on every call. Standard formats: HuggingFace, GGUF, TensorRT-LLM, ONNX, PyTorch state_dict.

Common workloads: Llama 3 instruct fine-tunes · Mistral Large licensed deployments · domain-specific embedding models · custom T5 / Whisper / BERT variants
AIR-GAPPED
BYOW · BRING YOUR OWN WEIGHTS

For weights that can never leave.

Same as BYOM, but the model file never traverses any TRD-controlled network. Weights are loaded from your storage volume directly into pod GPU memory. The runtime is fully air-gapped from inference. Even our operators cannot read the file — only the GPU sees it.

Fit for: defense contractors with ITAR-classified weights · biotech with trade-secret protein-folding models · sovereign-AI programs governed by national-security clauses
TRUST SUBSTRATE · INCLUDED

Same receipts. Same erasure.

A BYOM call emits the same ed25519-signed receipt as a TRD-native call. Inference duration, token counts, GPU hardware class, and carbon footprint all chain-anchor to Filecoin. GDPR Article 17 cryptographic erasure works identically — destroy the tenant DEK and every customer prompt + completion you stored becomes mathematically unrecoverable.

Verifiable: the receipt envelope identifies the model by SHA-256 of weights — auditors can prove a specific call ran the specific approved model version
Supported model formats
HuggingFace
safetensors · transformers
GGUF
llama.cpp · quantized 4/8-bit
TensorRT-LLM
NVIDIA engine plans
ONNX
cross-runtime portable
PyTorch
state_dict · TorchScript
vLLM
production serving
TGI
HuggingFace Text Generation Inference
Embedding-only
sentence-transformers · bge · gte
Model your stack
Pick GPU class + model size to see realistic deployment cost. Compare against your cloud baseline.
Open cost calculator →
09 — Sovereign as Substrate

Run the entire agent fleet inside your perimeter.

Sovereign isn't just a place to host inference. It's the substrate for everything TRD builds — the same Sovereign pod that runs your AI runtime can host the 641 industry agents from TRD Pilot. WhatsApp owner agents, agent-pay-agent commerce, the architect agent that builds sites — all of it inside your data perimeter, with the same signed-receipt + GDPR-erasure substrate.

AGENT FLEET

641 industry agents — local.

Restaurant booking agent, salon-stylist agent, visa-intake agent, healthcare-clinic agent. All 641 specialists across 87 industries run on the Sovereign pod. No traffic crosses your perimeter for agent invocations.

AGENT COMMERCE

Agent-pay-agent stays on-pod.

Agents transact via x402 micropayments without touching the public x402 network. Settlement happens against a pod-local USDC pool you control. Auditable, contained, signable receipts on both sides.

FEDERATION

Pods attest each other's chain heads.

Multi-region deployments (EU + UAE, say) federate via signed chain-head attestations. Each pod is independently authoritative in its perimeter; cross-pod auditors verify the full deployment without central TRD coordination.

FULL STACK

Your weights. Your agents. Your keys.

Bring fine-tuned Llama 3 or a licensed Mistral. Run the 641 TRD agents against it. Encrypt all receipts with a tenant-controlled DEK. Cryptographic erasure works the same way — destroy the DEK and every prompt, completion, and agent invocation is unrecoverable.

The full Sovereign-as-Substrate stack
Your GPUs Sovereign runtime Your weights (BYOM) 641 TRD agents Signed receipts GDPR cryptographic erasure
Six layers, one perimeter. No external API calls in the AI path. No data egress. The substrate every regulated AI deployment needs.
01 — Who it's for

Built for organizations that can't route data through external APIs.

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.

GOVERNMENT

Public sector

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.

FINANCIAL

Banks & insurers

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.

HEALTHCARE

Health systems

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.

02 — How deployment works

From scoping call to running pod, in a structured engagement.

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.

STAGE 01

Scope & audit

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
STAGE 02

Configure the pod

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
STAGE 03

Run inside your perimeter

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
03 — What's shipped & in flight

A foundation that's real today, not a pitch deck.

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.

sovereign.trdn.io marketing surface

LIVE

The public-facing surface explaining the self-hosted proposition to enterprise prospects — backend lead-capture API, frontend content, protected-zone enforcement. Shipped May 9.

Enterprise lead pipeline

LIVE

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.

6 architecture locks identified

LIVE

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.

Own-GPU inference engine

LIVE

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.

Speculative decoding

LIVE

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.

TRD Inference API · Tier 0

IN FLIGHT

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.

04 — What's coming

The roadmap to a repeatable enterprise deployment.

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.

PHASE 2 · IN FLIGHT

Architecture lock — design week

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.

Effort — 5 days, non-negotiable
PHASE 3 · UPCOMING

Deployment package — Docker / Kubernetes

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.

Effort — ~1 month
PHASE 3 · UPCOMING

Anthropic ZDR option — Tier -1

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.

Effort — 1–2 weeks
PHASE 4 · UPCOMING

First paying enterprise deployment

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.

Effort — ~1 month · founder-led
PHASE 5 · UPCOMING

SOC 2 · ISO 27001 · HIPAA certifications

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.

Effort — 2–3 months, on customer demand
PHASE 6 · UPCOMING

Multi-region pod automation

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.

Effort — 2–3 months
05 — vs. cloud AI APIs

Why TRD Sovereign outlasts the API your data can't touch.

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.

06 — Engagement model

Priced per deployment, scoped to your environment.

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.

STEP ONE

Infrastructure audit

Paid scoping

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.

  • 30-minute scoping call first
  • Full environment assessment
  • Jurisdiction gap analysis
  • Converts to contract deposit
ANNUAL CONTRACT

Sovereign deployment

$5k–50k / month

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.

  • Annual contract, per environment
  • 4-hour SLA support
  • Quarterly business reviews
  • Certification support on request

All figures reflect the engagement structure; exact scope and pricing are finalized in the infrastructure audit. The first deployments are founder-led.

07 — TCO Calculator

What does cryptographic sovereignty actually cost?

Model your full stack. Pick GPU class, model size, monthly token volume, and utilization. The calculator shows monthly cost across OpenAI, Azure OpenAI, AWS Bedrock, and TRD Sovereign — adjusted for your workload sensitivity tier. Numbers are public list prices as of May 2026; your specific quote may vary.

1M500M2B
20% (spiky)50%70% (typical)95% (saturated)
OPENAI · GPT-4O
$2,500
per month
⚠ Cannot deploy in sovereign regions
AZURE OPENAI
$3,375
per month
+35% regulated tier addon
AWS BEDROCK · CLAUDE
$1,000
per month
+25% regulated tier addon
RECOMMENDED
TRD SOVEREIGN · SELF-HOST
$650
per month · amortized
All compliance built in. 81% vs Azure.
🌱
CARBON RECEIPT · MONTHLY EMISSIONS
0 kg CO₂e with verifiable receipt · cloud vendors: 0 kg CO₂e, no receipt
Receipt-backed
✓ C2PA + Filecoin
ANNUAL SAVINGS vs AZURE
$32,700
Cumulative over 12 months · plus compliance posture · plus carbon receipts
Get a custom quote →
Typical deployment: 4-8 weeks · Scope per environment
3-YEAR CUMULATIVE TCO
Linear monthly accrual at current settings · upfront deployment cost amortized over 36 months
Azure AWS TRD Sovereign
$0 M0 12mo 24mo 36mo
Methodology & assumptions

Pricing baselines (May 2026 list prices):

  • OpenAI GPT-4o: $2.50/M input + $10/M output → ~$25/M blended (60/40 in/out)
  • Azure OpenAI: same per-token as OpenAI + 35% regulated tier surcharge (private endpoint, BAA, regional residency)
  • AWS Bedrock Claude Sonnet: $3/M input + $15/M output → ~$8/M blended + 25% compliance addon
  • TRD Sovereign: amortized GPU cost (H100 ~$2.50/hr or equivalent in-region) at typical 70% utilization for 8B-class models → ~$0.50-1/M blended

Sensitivity tiers:

  • Standard: public-cloud cost as-is. Sovereign self-host: ~3x cheaper for >100M tokens/mo.
  • Regulated: compliance addons applied to public cloud. Sovereign self-host: same baseline (compliance built in).
  • Sovereign-Only: public cloud cannot deploy (regulatory bar). Only sovereign self-host is viable.

Numbers exclude one-time deployment scoping ($5-50k typical) and exclude license/support fees. Final quote may differ based on environment, GPU class, redundancy SLA, and integration scope.

07 — Compliance posture

Certifications pursued per deployment, not for show.

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.

SOC 2 Type 2
Pursued on customer demand
ISO 27001
Pursued on customer demand
HIPAA
Pursued on customer demand
GDPR alignment
Architecture aligned
UAE data residency
In-region deployment
India DPDP
In-region deployment

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.

08 — Common questions

The questions regulated buyers actually ask.

No. Inference, orchestration, storage, and observability all run inside your perimeter. The only outbound connection is an optional license-server callback that verifies model weights — it transmits a license token and nothing else. For fully air-gapped deployments, even that can be handled offline.
The default is Qwen3-32B-FP8 served via vLLM — chosen for the quality-to-speed sweet spot, FP8 quantization that fits comfortably on a single H200, strong coding and reasoning performance, and open weights with no licensing concerns. The model choice is part of your deployment config — it's your decision, set when the pod is configured.
The own-GPU inference layer benchmarked at 1,605 tokens/second aggregate at concurrency 16 on real production workloads, with speculative decoding contributing a 25%+ throughput gain (66% draft-token acceptance). Exact numbers depend on your hardware tier and concurrency — the infrastructure audit scopes this for your environment.
A raw open-source model is an endpoint. TRD Sovereign is the full agent stack — all 641 agents, the complete build pipeline, the memory layer, the orchestrator — packaged to deploy with Helm or Compose and run with zero external calls. You're not assembling infrastructure; you're deploying a finished, observable, supported system.
Yes. The pod is designed with no outbound dependencies beyond the optional license callback, and that can be satisfied offline for air-gapped environments. Public-sector and defense deployments are a core use case, not an afterthought.
It starts with a 30-minute scoping call, then a paid pre-deployment infrastructure audit (the fee converts to a deposit on a signed contract). From there: a 30-day proof of concept at a flat $5k, then an annual contract — $5k–50k/month depending on scale — including deployment, training, 4-hour SLA support, and quarterly reviews. The first deployments are founder-led.
You do. The model weights are deployed into your infrastructure, and all keys are yours — managed per-tenant via the encryption envelope locked in the Phase 2 architecture. TRD does not hold a copy of your keys and has no path to your data, by design.
That's the upcoming Anthropic Zero Data Retention tier — a middle path where request data is never stored by the provider. It's premium-quality inference with stronger privacy than a standard API, for organizations whose constraints allow it. It's on the Phase 3 roadmap, not yet live.
The pod's architecture — no egress, your keys, your infrastructure, your region — maps cleanly onto most data-residency and sovereignty regimes. We pursue formal certification when a specific deployment requires it, and we'll share a gap analysis for your jurisdiction before you commit to anything.

Talk to us about your environment.

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.

130-minute scoping call — environment, hardware, compliance needs.
2Paid pre-deployment infrastructure audit, if it's a fit.
3Audit fee converts to deposit on signed contract.