Confidential AI Agents on TEE Models in Intel TDX

Why confidential agents exist

Regulated work does not stop because ChatGPT is convenient. Lawyers still draft contracts under attorney-client privilege. CFOs still build forecasts on material non-public information. Hospitals still process protected health information under HIPAA and the GDPR Article 9 special-category rules. Tax teams still build transfer pricing files that are statutory-privileged under §6713. For every one of these workloads, the moment data leaves your perimeter and lands inside a third-party model the legal posture changes, usually for the worse.

Intel TDX (Trust Domain Extensions) runs the model inside a hardware enclave with CPU-level memory encryption: the host operating system of the GPU machine sees ciphertext, not your prompt. On the way there, our servers parse each prompt and upload in clear and relay it over TLS; nothing is used for training. GET /v1/attestation returns the Intel TDX quote and the NVIDIA GPU evidence of the instances serving the model, on a nonce you choose, so your compliance team can check that the model runs inside a TDX domain, not on a soft-isolated VM with the same marketing copy. It covers the serving instances, not that a given response came from them.

VOLTAGE EI, a French company, is the controller, and the customer database sits in Frankfurt. The compute itself runs on capacity operated by our listed sub-processors in the US, under Standard Contractual Clauses, protected by Intel TDX rather than by location, and that transfer is documented for your Data Protection Impact Assessment. Every agent below runs on the same architecture: the only thing that differs is the system prompt, the document schema, the regulatory framework it cites and the buyer it was designed for.

None of these agents replace a partner, a CPA, a clinician or a CISO. They replace the analyst hours that come before partner sign-off. They are framed and priced accordingly, $20/month for personal use, $349/month for small teams, $1,199/month for departments, custom for enterprise rollouts with SSO/SAML, signed DPA, dedicated TDX capacity and fine-tuning on your playbook.

The 9 agents

Each agent has its own system prompt, regulatory framework, document schema and output structure. Click into any one to see capabilities, workflow, example outputs, pricing and FAQ.

How to choose

Pick the agent closest to the regulatory framework that already governs your work. The Contract Analyst is the right starting point for any team where documents are protected by professional secrecy. The Financial Analyst is built for MNPI workflows. The Compliance Officer covers multi-framework gap analysis. The Medical Records Analyst is the HIPAA / GDPR Art. 9 path. Due Diligence is the M&A path. Cybersecurity is the SOC and CVE triage path. HR covers pay equity, RIF planning and investigations. Tax covers transfer pricing, Pillar Two and provision review. Teams that need an agent we have not built can deploy any CrewAI, LangChain or OpenClaw agent on the same confidential infrastructure via API. Developers whose own coding agent needs a confidential GPU can start with the agent quickstart.