Due Diligence AI Agent,
M&A Document Review in TDX.
Batch-review deal documents with an AI model following a due-diligence checklist, with inference in Intel TDX. Structured risk flags and extracted terms per document. Outputs are not legal advice.
From Starter $349/month. Up to 20 files per bulk request.
The pain
Data rooms break review tools and weekend plans.
Large rooms, finite associates
Deal rooms routinely run to thousands of files. Associates spend weekends paginating through duplicate NDAs and out-of-scope finance schedules, and material risks can slip through.
Confidentiality is usually contractual
Without hardware isolation, the confidentiality of target documents rests on contract terms and the provider's access controls. Here, inference runs in Intel TDX and we don't store your prompts.
Multilingual, messy files
A French target with a German subsidiary register and English material contracts is common. The agent reads EN, FR, DE, ES and IT. Image files (PNG, JPG, WebP) are OCRed on upload; scanned, image-only PDFs need OCR before upload.
Workflow
Five steps from VDR to deal memo.
Ingest
Upload documents to the Due Diligence agent in chat (PDF, DOCX, TXT, CSV, images), or send up to 20 PDF, TXT or CSV files per request to the bulk analysis endpoint. There are no VDR connectors. Documents pass through our servers to a TDX enclave and are not kept.
Categorize
Each bulk result names the document type. In chat, ask the agent to sort documents into a deal-room taxonomy: corporate, contracts, IP, employment, real estate, finance, regulatory, litigation, data protection.
Extract
The bulk endpoint extracts parties, amounts, dates, durations and clauses, and runs a second pass that flags values that differ for human review. In chat, ask for specific fields such as change-of-control or key-person clauses.
Score
Each document gets an overall grade (CRITICAL / HIGH / MEDIUM / LOW) and risk flags, each with a clause or section reference, the issue and a recommendation.
Summarize
Each result includes a short summary and a recommended next step. Ask the agent in chat to compile findings into a table or memo, then export the conversation to Markdown. There is no Excel or DOCX export.
Supported document categories
Built around the deal-team taxonomy.
- · Articles
- · Cap table
- · Shareholder agreements
- · Board minutes
- · Subsidiary register
- · Top customers
- · Top suppliers
- · JVs
- · Distribution
- · License-in / license-out
- · Patents
- · Trademarks
- · Domain register
- · Source code license inventory
- · Open-source compliance
- · Key-person contracts
- · Restrictive covenants
- · Bonus plans
- · Stock options
- · Convention collective
- · Lease register
- · Termination notices
- · Environmental reports
- · CapEx commitments
- · Pending cases
- · Settled disputes
- · Regulatory inquiries
- · Insurance claims
- · DPIAs
- · RoPA
- · Sub-processor list
- · Breach history
- · CNIL correspondence
- · Audited statements
- · Mgmt accounts
- · Net debt schedule
- · CapEx forecast
- · Working capital
Output
Risk flags, summary, extracted terms.
What the bulk endpoint returns for each document, as structured JSON. In chat, ask the agent to compile findings across documents and export the conversation to Markdown. There is no Excel or DOCX export.
Overall grade · flags with clause reference, issue, recommendation
Two to three sentences per document · recommended next step
Parties, amounts, dates, durations, clauses · values that differ between two passes flagged
Pricing
Starter
$349
2,000 requests / mo · 3 seats
Pro
$1,199
5,000 requests / mo · DeepSeek-V3.2-TEE
Enterprise
$3,499
Unlimited requests · SCIM · RBAC
Plus plan ($20/mo) covers single-document review in chat.
FAQ
Frequently asked questions.
How does the agent decide what is CRITICAL vs HIGH?
It is an AI model following a review checklist, so each grade is the model's judgment, not a calibrated score. There is no playbook upload yet: paste your own thresholds in chat (for example, change-of-control without consent on a top-10 customer = CRITICAL) and the agent applies them as instructions.
What about non-English documents?
The agent reads EN, FR, DE, ES and IT. A French M&A data room with a German subsidiary register and English material contracts is the typical case: the agent reads each in its own language and answers in the deal-team working language.
Can findings link back to the source document?
Each bulk result carries the file name, and each flag a clause or section reference taken from the text. There are no page numbers and no links into a VDR.
Is the data room shared with the underlying LLM provider?
Inference runs on a third-party TEE inference network, named as a sub-processor in our DPA. Documents pass through our servers to a TDX enclave and are not kept. Enclave memory is encrypted by the CPU and the host's hypervisor cannot read it; each request runs on one attested TEE instance. The open model weights are frozen and are not trained on your documents.
Does the data room persist across the deal?
Only if you file documents into a Matter (Beta) at /workspaces: they are stored as notes, encrypted at rest, in the EU. The assistant does not read or cite Matter files yet. Starter includes 10 Matters, Pro 25, Enterprise unlimited.
Can we watch a data room for new uploads?
No. The data-room watcher is disabled, and scheduled tasks read no files and check no live source.
Which VDRs and DMS does the agent connect to?
None. There are no Datasite, Intralinks, Ansarada, iDeals, Drooms, iManage or NetDocuments connectors. Connectors at /connectors cover Gmail, Google Drive, Slack, GitHub and Microsoft 365. Starter includes 3 connectors, Pro 5, Enterprise unlimited.
Run a pilot on a small deal first.
Start with a single bilateral deal on Starter. Scale to mid-market on Pro when you're ready.