CIM, QoE, financial model and management presentation review without sending MNPI to OpenAI, Anthropic or Google.
Every CIM, financial model, QoE report and management presentation contains Material Non-Public Information. Sending it to ChatGPT or Claude can be a securities-law and NDA problem. Here, documents are parsed by our servers and passed to an open-weight model in an Intel TDX enclave; never used for training. The agent cross-references documents for inconsistencies, scrutinizes EBITDA add-backs, scores deal risk across 8 dimensions and generates the IC summary. Built for deal teams running confirmatory DD.
Built for: PE/VC associates and partners, corporate development teams, M&A bankers, investment committees
DD data rooms run to hundreds or thousands of documents, and analyst teams spend hundreds of hours per deal on document review. A single missed red flag, recurring "one-time" add-backs, undisclosed related-party fees, customer concentration in expiring contracts, can cost far more than the DD budget after closing. Yet most AI tools process MNPI on shared US infrastructure.
Turn days of analyst document review into a structured cross-document risk register and IC-ready summary, with inference in an Intel TDX enclave and nothing stored.
What the Due Diligence Analyst does on every document, with a TEE model that runs inside an Intel TDX hardware enclave.
The single most important QoE work. Calculates add-back ratio explicitly, flags add-backs >30% (RED) or >50% (BLACK), surfaces "one-time" charges that recur across periods, and notes which add-backs a PE buyer typically refuses to credit.
Upload CIM + financials + management presentation. The agent flags where revenue figures, employee counts, customer counts or growth claims disagree across documents, the highest-value finding in any DD process.
Financial, legal/regulatory, operational, market/commercial, technology/IP, people/culture, regulatory/antitrust, integration, each scored 1-5 with specific findings and mitigation options (price adjustment, escrow, earnout, R&W, indemnification).
Decomposes revenue into recurring / repeating / re-occurring / one-time, surfaces customer concentration, NRR/NDR trend lines and revenue-mix shifts that degrade quality even when total revenue grows.
Compares the asking multiple against precedent transactions and public comps. Flags whether the price reflects the identified risks or assumes a clean balance sheet that does not exist.
One-page Investment Committee output: recommendation (PROCEED / PROCEED WITH CONDITIONS / PAUSE / PASS), three reasons in, three reasons out, key protections required, and a 100-day integration risk read.
Four steps from upload to export. Your document is parsed by our servers and passed to a TEE model; never used for training.
CIM, teaser, management presentation, QoE report, financial model (as CSV), legal DD memo. PDF, DOCX and CSV files go over TLS, are parsed by our servers and passed to a TEE model; never used for training.
The model runs inside an Intel TDX hardware enclave, with memory encrypted by the CPU during inference. Nothing is sent to OpenAI, Anthropic or Google. Check your NDA terms before uploading.
Deal summary, EBITDA bridge, revenue quality matrix, 8-dimension risk scorecard, cross-document inconsistency table, valuation sanity check, 10-15 DD questions for management.
Markdown for the IC memo, tables you can paste into Excel for the deal model, or the OpenAI-compatible API (Pro tier) for your own deal-CRM scripts.
Every output cites the specific article, section or control. The Intel TDX hardware boundary handles the data-protection side; the agent handles the analytical side.
Real prompts from investment / m&a teams.
Analyze this CIM, what's the investment thesis and what are the red flags?
Scrutinize the EBITDA add-backs in this QoE report. Which ones are legitimate?
Score this deal across all 8 risk dimensions. What should the IC know?
Cross-reference the CIM revenue claims against these actual financial statements.
Prompt: Scrutinize the EBITDA add-backs in this QoE.
Output: Adjusted EBITDA $14.2M vs reported EBITDA $8.6M. Add-backs = $5.6M (39% of adjusted EBITDA, RED). Concerns: (1) "One-time" CEO bonus add-back has appeared in 3 of last 4 years, recurring, not one-time; (2) $1.4M "synergy run-rate" add-back is forward-looking, not historical, most PE buyers reject; (3) $0.9M legal settlement add-back has documented follow-on litigation pending. Defensible add-backs total $2.1M → defensible Adjusted EBITDA = $10.7M.
Prompt: What questions should we ask management in the next DD session?
Output: 1. Top-5 customer concentration jumped from 28% to 41% YoY, what is the renewal timeline and price-increase exposure? 2. Gross margin compressed 320bp; which COGS line caused it? 3. CIM claims 94% NRR but financial supplement shows 87%, reconcile. 4. Three "one-time" restructuring charges in three years, what is the underlying operational issue? 5. Related-party management fee of $480K, what services and would they survive change of control?
Every tier runs inside Intel TDX hardware enclaves. Plans stay in sync with /pricing.
1,000 requests/month, 1 seat. For independent operators, fractional CFOs, search-fund principals.
3 seats, 2,000 requests/month, 100 MB uploads. For lower mid-market PE and corporate development teams.
10 seats, 5,000 requests/month, 500 MB uploads, API access for deal-CRM integration, 12-month audit log retention.
Unlimited seats, fine-tuning on your firm’s deal playbook, SSO/SAML, dedicated TDX capacity, custom NDA, signed DPA.
Honest comparison. Hardware-rooted confidentiality is what most alternatives are missing.
| Alternative | Pros | Cons vs VoltageGPU |
|---|---|---|
| Hebbia |
|
|
| Rogo |
|
|
| Internal analyst team + ChatGPT |
|
|
That depends on your NDA, and it is your call. Most NDAs restrict disclosure to third parties and limit use to evaluating the deal. What we can state: documents are parsed by our servers and passed to an open-weight model in an Intel TDX enclave, never used for training, and nothing goes to OpenAI, Anthropic or Google. We can sign a per-deal NDA on the Enterprise tier.
Documents are parsed by our servers and passed to the TEE model; never used for training. During inference, Intel TDX encrypts the model memory in hardware, and nothing is sent to OpenAI, Anthropic or Google. This reduces third-party exposure; it does not replace your own information-barrier and NDA controls.
The default TEE model of our catalog inside an Intel TDX enclave on the Free tier. Plus, Team Starter and Pro tiers use the larger Qwen3.5-397B-A17B (256K context, useful for whole-CIM analysis). Pro and Enterprise can route to DeepSeek-V3.2-TEE reasoning for valuation scenario work.
No. It takes on first-pass document reading and structured extraction. Judgment calls, deal pricing, negotiation strategy, IC vote, remain with the deal team, and every finding should be checked against the source.
Yes. The model handles English, French and German. For French SAS structures, it understands BSA, BSPCE, and ratchets, flagging clauses that hurt common holders. For German GmbH / GmbH & Co. KG structures, it handles the Vorbehaltsgut and Sicherheiten complexity.
Not natively. On the Pro tier, the OpenAI-compatible API lets you script things like CIM summaries into deal-CRM notes, red-flag screens on inbound teasers or IC memo drafts. Requests pass through our relay, which sees them in clear (TLS ends at our host). To keep plaintext off our servers, run your own inference server in a Confidential VM and terminate TLS inside its trust domain.
The Financial Analyst agent is a better fit for ongoing portfolio monitoring. The Due Diligence Analyst is optimized for pre-close deal evaluation; the Financial Analyst handles monthly board pack analysis and covenant tracking.
External DD support (Big Four advisory) runs $300-800/hr. The Pro tier ($1,199/mo) includes 5,000 requests/month. The agent supplements rather than replaces the QoE provider; it shortens first-pass reading.
Model in an Intel TDX enclave, with an attestation of its serving instances on your nonce. French company (VOLTAGE EI), compute at US sub-processors under SCCs. Cancel anytime.
