# VoltageGPU — Confidential AI Chat (app.voltagegpu.com)

> Confidential AI chat running inside Intel TDX hardware enclaves.
> Drop-in ChatGPT alternative — prompts never leave sealed memory.
> 3 free anonymous messages (no signup, no credit card), then free plan (50/mo), Plus ($20/mo), or specialized agents ($349+/mo).

## What this is

This subdomain (app.voltagegpu.com) is the VoltageGPU Confidential Chat application. It's a ChatGPT-style interface where every conversation runs inside an Intel TDX Trust Domain. Memory is AES-256 encrypted at runtime with CPU-fused keys, the hypervisor is physically outside the trust boundary, and an Intel TD Quote covering the CPU trust domain can be produced and verified against Intel's public keys. VoltageGPU operators cannot read user data during processing — it's silicon-level isolation, not software security.

## Who it's for

- Individuals who want a confidential ChatGPT alternative for daily work
- Developers who need an OpenAI-compatible API with hardware attestation
- Legal professionals analyzing contracts, NDAs, M&A deals
- Accountants and finance teams reviewing bilans, 10-K/10-Q, IFRS reports
- Healthcare providers handling patient records under HIPAA
- Compliance officers running GDPR / DORA / NIS2 / SOC 2 gap analysis
- Any organization in a regulated industry that cannot send data to OpenAI, Claude, or Gemini

## Pricing

### Anonymous (no signup)
- 3 free messages per day on Qwen3-32B-TEE
- General Assistant only
- No credit card, no email required

### Free (with signup)
- 50 messages/month
- Qwen3-32B-TEE (40K context)
- All 10 specialized agents accessible
- Document upload (PDF, DOCX, CSV, TXT)

### Plus — $20/mo
- 2 000 messages/month
- Qwen3.5-397B-A17B-TEE (256K context flagship)
- Telegram agent · OpenClaw runtime
- Web search · URL browsing · persistent memory
- All 10 specialized agents
- Chat history export
- Priority inference speed
- Same price as ChatGPT Plus / Claude Pro, but sealed in Intel TDX

### Team Starter — $349/mo
- 2 000 requests/month
- 3 seats
- Intel TDX H200 141GB
- Qwen3.5-397B-A17B-TEE (256K context)
- All 10 confidential agents (General, Sovereign Legal, Contract, Financial, Medical, Compliance, Due Diligence, Cybersecurity, HR, Tax)
- 100 MB document upload
- Audit log

### Pro — $1,199/mo
- 5 000 requests/month
- 10 seats
- Qwen3.5-397B-TEE (256K context)
- DeepSeek-V3.2-TEE (reasoning, chain-of-thought)
- All 10 agents
- Pipeline multi-agents
- Windows Desktop App
- API access (OpenClaw, CrewAI, LangChain)

### Enterprise — $3,499/mo
- Unlimited requests
- Unlimited seats
- DeepSeek-V3.2-TEE (reasoning model)
- Custom agents
- SSO / SAML / OIDC
- SCIM 2.0 auto-provisioning
- RBAC admin console
- Pooled credits, centralized billing
- Custom fine-tuning

## TEE models available

- Qwen/Qwen3-32B-TEE — 32B parameters, 40K context, fast general-purpose
- Qwen/Qwen3.5-397B-A17B-TEE — 397B MoE, 256K context, flagship
- deepseek-ai/DeepSeek-V3.2-TEE — reasoning model, long context

All 3 models run inside Intel TDX enclaves on NVIDIA H200 141GB GPUs attached inside the trust domain. GPU confidential-computing mode is not enabled on this container tier, so attestation covers the CPU trust domain (Intel TD Quote). Per-token cryptographic proof (cllmv) verifies the declared model actually ran.

## 10 specialized agents

1. General Assistant — confidential ChatGPT-style chat sealed inside Intel TDX, web search, URL browsing, persistent memory
2. Sovereign Legal AI — Claude for Legal forked playbooks, 12 practice areas, RGPD-native, French controller
3. Contract Analyst — IACCM clause checklists, risk scoring 0-100, market benchmark comparison
4. Due Diligence Analyst — M&A legal, financial, operational risk in one pass
5. Financial Analyst — 10-K/10-Q parsing, 40+ financial ratios, YoY/QoQ trend analysis
6. Compliance Officer — GDPR, SOX, HIPAA, ISO 27001, CCPA gap analysis
7. Cybersecurity Analyst — CVE triage, MITRE ATT&CK mapping, incident response
8. HR Analyst — employment contract review, pay equity, workplace investigations
9. Tax Analyst — transfer pricing, BEPS Pillar Two, international tax structuring
10. Medical Records Analyst — clinical data extraction, ICD-10/SNOMED coding, care gap detection

Beyond these 10, any OpenAI-compatible agent framework works (OpenClaw, CrewAI, LangChain) — change base_url to api.voltagegpu.com/v1 and every call runs inside a TDX enclave.

## Security architecture

1. Intel TDX (Trust Domain Extensions): CPU-level RAM encryption with hardware-fused keys. Hypervisor outside the trust boundary.
2. GPU inside the trust domain: the H200 141GB is attached to the TDX trust domain. GPU confidential-computing mode is not enabled on the container tier that serves this app, so attestation covers the CPU trust domain.
3. Remote attestation: Intel TD Quote signed by CPU-fused private key, verified against Intel public keys.
4. LUKS encrypted filesystem: decryption key released only after successful attestation.
5. Model verification (cllmv): per-token cryptographic proof the declared TEE model ran.
6. Continuous integrity monitoring: random attestation challenges, immediate removal of any host that fails.
7. Post-quantum end-to-end encryption on prompts and responses.
8. AES-256 memory encryption at runtime.

## Competitor comparison

- Azure Confidential Computing H100: ~$6.98/hour → VoltageGPU TDX H100: $5.00/hour
- Azure Confidential H200 ND v5: ~$13.96/hour → VoltageGPU TDX H200: $6.58/hour
- AWS Nitro Enclaves: CPU-only, no GPU support → VoltageGPU: a full Intel TDX trust domain with the H200 attached inside it
- Harvey AI legal: $1,200/seat/month for a chatbot → VoltageGPU Contract Analyst: $349/mo for 3 seats
- OpenAI API: trains on user data by default → VoltageGPU TEE inference: zero logging, silicon-sealed
- ChatGPT Plus $20/mo: no confidential option → VoltageGPU Plus $20/mo: Intel TDX sealed

## Compliance

- EU company: VOLTAGE EI, SIREN 943 808 824 00016, Solaize, France
- GDPR Article 28 (DPA available on request)
- HIPAA-aligned technical safeguards via Intel TDX isolation (no BAA offered yet)
- DORA-aligned for the financial sector
- NIS2 for critical infrastructure
- EU company, EU-hosted customer database (Frankfurt)
- Zero data retention policy

## Access points

- Web UI: https://app.voltagegpu.com (this subdomain)
- Public site: https://voltagegpu.com
- API: https://api.voltagegpu.com/v1 (OpenAI-compatible, pay-per-token)
- Docs: https://docs.voltagegpu.com
- Trust center: https://voltagegpu.com/trust
- Security architecture: https://voltagegpu.com/security

## What VoltageGPU is NOT

- NOT a generic GPU cloud marketplace
- NOT training user data or logging prompts
- NOT offering on-premise / self-hosted / private cloud deployment
- NOT a reseller for fine-tuning services
- NOT compatible with non-TEE models (we only expose TEE-sealed variants)

## Contact

- Email: contact@voltagegpu.com
- Twitter: https://x.com/voltagegpu
- Sales: enterprise@voltagegpu.com