VoltageGPU ships three confidential AI agents for finance teams that have to document where ledgers, AML alerts and audit work-papers go: an AML Monitoring agent, an Audit Automation agent and an FP&A agent. Inference runs in Intel TDX on a third-party TEE inference network, with memory encrypted by the CPU. We don't store prompts, transaction data or forecasts. EU company (VOLTAGE EI, France, SIREN 943 808 824), native RGPD Article 28 Data Processing Agreement.
Microsoft Copilot for Finance and ChatGPT Enterprise come from US companies (Microsoft, OpenAI). Both are excellent products, with no published TEE inference. VoltageGPU runs open-weights models in Intel TDX on a third-party TEE inference network, does not store prompts, and names its sub-processors.
Starter ($349/mo), 3 seats, all 10 confidential agents on Qwen3.5-397B-TEE (256K context). Pro ($1,199/mo), 10 seats, adds DeepSeek-V3.2-TEE reasoning and the OpenAI-compatible API. Enterprise ($3,499/mo), unlimited seats, SCIM 2.0, RBAC admin console, admin audit log.
AML monitoring, audit automation, FP&A. Inference runs in Intel TDX. EU company, EU-hosted customer database. Your chats stay in your browser, not on our servers.
Named sub-processors and a plain data-flow description for your DORA Article 28-30 register. We don't store prompts, so there is no prompt archive to hand over.
The financial-data confidentiality problem
CFOs, Chief Compliance Officers and Heads of Internal Audit are asked the same question: where is the AI that lets us close books faster, triage AML alerts faster and draft audit work-papers faster, without breaking DORA, GDPR or SOX? The default answers, Microsoft Copilot for Finance, ChatGPT Enterprise, Google Gemini, come from US companies running their own cloud infrastructure.
For an EU regulated financial entity, GDPR Article 32 (security of processing) and DORA Article 28-30 (third-party ICT risk) mean the DPO has to document where that data goes, who processes it and under which contract, before a French ACPR or German BaFin examiner asks.
VoltageGPU keeps that answer short. We are an EU company (VOLTAGE EI, France, SIREN 943 808 824) with a named sub-processor list. We run Qwen3.5-397B and DeepSeek-V3.2 (open weights) and, on the Free plan, a TEE model from our catalog, on a third-party TEE inference network inside Intel TDX, with memory encrypted by the CPU. Prompts reach our API over TLS; we relay them to the TDX enclave without storing them.
Three confidential finance agents
Each agent is an AI model following a review checklist for one finance workflow. None of them are regulated advisors, they are decision-support tools that a qualified compliance officer, auditor or finance professional reviews and signs.
AML Monitoring
Reviews transaction batches against FATF / 6AMLD typologies, drafts SAR-ready narratives and flags structuring, layering and unusual cross-border patterns for a human MLRO to sign.
Audit Automation
Drafts walkthroughs, sample-test summaries, control narratives and audit work-papers. Useful for both internal audit functions and external auditors.
FP&A
Variance analysis, forecast drafting, board memo drafting and sensitivity-table generation on management accounts you would rather not paste into a general-purpose chatbot.
DORA mapping
The Digital Operational Resilience Act applies to banks, insurers, investment firms, crypto-asset service providers and most EU financial entities. We provide the information regulated buyers need to assess VoltageGPU as an ICT third-party provider.
DORA Art. 28
Third-party ICT risk register
VoltageGPU appears as an EU-based ICT third-party provider. We supply the register fields (entity, jurisdiction, sub-processors, data flow, exit plan) on request.
DORA Art. 29
Concentration risk assessment
Sub-processors are named in the DPA, including the third-party TEE inference network, so you can assess concentration risk yourself.
DORA Art. 30
Contractual provisions
Article 30 requires terms on processing locations, audit rights, exit and incident notification. We review them with you against our DPA; inference runs on a third-party network, not only in the EU.
DORA Art. 17
ICT-related incident reporting
No SIEM or webhook integration today. Incident classification under the DORA technical standards stays with you.
SOX angle
SOX Section 404 internal control testing produces work-papers, walkthroughs and substantive test evidence that auditors must protect under PCAOB independence and confidentiality requirements. Any AI tool that sees that material becomes a sub-processor the firm has to approve. Pasting it into Microsoft Copilot inside a client's Microsoft 365 tenant blurs the line between auditor data and auditee data.
VoltageGPU keeps the data flow short. Prompts reach our API over TLS; we relay them to a third-party Intel TDX inference network without storing them, and chats stay in the auditor's browser. Each request runs on one attested TEE instance.
We position VoltageGPU as audit tooling. The auditor's professional judgment, sample selection and conclusions remain the auditor's. The agent drafts; the human signs.
VoltageGPU vs Microsoft Copilot for Finance vs ChatGPT Enterprise
We have nothing against Copilot or ChatGPT Enterprise, they are excellent products. The difference is where inference runs and what is stored.
Pricing
Flat monthly tiers. No per-token surprises. Inference quotas reset every month. Annual billing saves 20%.
Starter
$349/ mo
Pro
$1,199/ mo
Enterprise
$3,499/ mo
FAQ
Can a regulated bank legally use this under DORA?
That depends on your own DORA risk assessment, register entry and board approval for ICT third parties. We provide the EU counterparty details, the named sub-processor list and a plain description of the data flow, including that inference runs on a third-party TEE network.
Is anything sent to OpenAI, Anthropic or any US provider?
Nothing goes to OpenAI or Anthropic. The agents run Qwen3.5-397B and DeepSeek-V3.2 (open weights) and, on the Free plan, a TEE model from our catalog, on a third-party TEE inference network inside Intel TDX. That network is named in our sub-processor list; we do not claim it sits outside US jurisdiction.
Is this regulated financial advice?
No. VoltageGPU is a confidential AI tooling platform. The agents draft suggestions a qualified MLRO, internal auditor or finance professional reviews and signs. We do not offer regulated financial, legal or audit advice.
How does the Audit Automation agent stay independent of the auditee?
Chats stay in the auditor's browser and we don't store prompts or documents, so work-papers are not kept on our side. Documenting AI tool usage under PCAOB AS 1215 stays with the engagement team.
What if our DPO requires SOC 2 or ISO 27001 before signing?
We hold no SOC 2 or ISO 27001 certification today.
Can the agents run on cron, before the team logs in?
Only as a custom scheduled prompt at /scheduled. Scheduled runs have no access to the web, your files or connected systems, so they cannot run sweeps, re-score alerts or collect evidence.
Does it pull from our GL, TMS, case manager and DMS?
No. Connectors at /connectors cover Gmail, Google Drive, Slack, GitHub and Microsoft 365. There is no GL, TMS, case-manager or sanctions-feed connector. Starter includes 3 connectors, Pro 5, Enterprise unlimited.
Keep going
Sovereign suspicious transaction triage. FATF and 6AMLD aligned. SAR-ready drafts.
Walkthroughs, sample-test summaries, SOX 404 control narratives, inference in Intel TDX.
Variance analysis, forecast drafts, board memos. Inference in Intel TDX.
The hub: vertical agents for legal, finance and regulated professionals.
How VoltageGPU lines up with DORA Article 28-30 third-party risk obligations.
Article 26 deployer evidence, model cards, logging, under the staged 2026 timeline.
Plus / Starter / Pro / Enterprise, see what each tier includes.
How requests reach the Intel TDX enclave, and what we store.
The flagship legal agent. Useful reference for the same architecture pattern.
Sector page: AML, KYC, DORA and SOX-aligned agent workflows.
The full agent line-up across legal, finance, healthcare and operations.
The infrastructure layer: Intel TDX confidential compute, as raw compute.
Pick a tier, review our DPA and sub-processor list with your DPO, and try the agents on real work. Or talk to us first, we answer DPO and CCO questions directly.