The data room that never takes your files.

A virtual data room is a secure online space where a company shares confidential documents with investors, acquirers or licensing partners during due diligence. Classic VDRs store copies of the documents and control who can view them. Haihtuva works differently: the seller's documents stay in its own storage (Google Drive, Microsoft OneDrive or SharePoint, Dropbox) and are read on demand, read-only.

Haihtuva is built AI-first rather than adding AI to a file-sharing product. Each buyer runs its own AI due diligence on the shared documents inside a sealed Azure confidential virtual machine in the EU. Documents are processed in encrypted memory and are not copied to disk, and when the room closes its data and keys are deleted.

AI due diligence uses language models to read the documents of a deal, find risks with page-level citations, answer questions from the data room and draft reports. In Haihtuva every finding cites the document and page it comes from, so people can check it. AI output is a draft to verify; it is not legal, financial or investment advice.

Yes. In Readiness the seller links its storage and sees its data room the way a buyer will: a tailored checklist, missing documents, likely buyer questions and drafts of missing documents from its own files. Buyers never see the seller's preparation.

Each buyer works in a private lane: other bidders never see its questions, settings or reports. The seller sees which documents each organisation opened, not per-person timings. The operator, BSUM, has no application access to the documents and receives only technical health data, never document content.

Equity investments, licensing and partnering, and M&A in any industry. The analysis has industry profiles for life science, software and IT, energy, manufacturing, financial services and real estate, and a general profile for everything else.

Yes. Each buyer chooses its AI model (the sealed in-room model or its own model account), edits the instructions the AI follows for every step, chooses the report sections and how much runs by itself, and adds its own checklists, question lists and training material. These settings stay confidential in the buyer's private review space: the seller, other bidders and BSUM never see them. A change of model or provider is approved by both sides in the room's security summary before it is used.

No. BSUM does not use room content, questions or results to train AI models, and configures the model providers it selects not to do so. The approved model and its settings are listed in each room's Security Specification.

Not yet. A SOC 2 readiness programme has been initiated; an independent audit has not yet been performed. Our security overview and the Security Specification of each room describe the controls in place today.

Google Drive, Microsoft OneDrive and SharePoint, Dropbox and Box, read-only; files stay where they are. If your organisation does not approve a storage connection, you can upload files, folders or a zip instead: they are kept in the room's encrypted memory only.

PDF (including scans, read with OCR in about 40 languages), Word (.docx and .doc), Excel (.xlsx and .xls, with cell references and arithmetic checks), PowerPoint (.pptx and .ppt), OpenDocument, RTF, text, images, and video and audio: speech is transcribed inside the room by an open model installed with it, and on-screen text is read from the frames. Nothing is sent to an outside transcription service.

Models provided by BSUM, with no account of your own: the sealed open model and premium models hosted on BSUM's Microsoft Azure (for example GPT-5.5, Claude and DeepSeek). Each room's security summary names the model, its host and region before both sides approve. Claude models are processed in the United States, and the summary says so. Or connect your organisation's own approved model account.

Recovery saves are on by default: the room's own records (settings, users, Q&A, reports, audit trail) are saved as an encrypted copy that only the room's attested machine can open, so a restart resumes where it stopped. Documents are never in it: they are read again from your storage. Storage sign-ins and your own AI keys are never saved. The seller admin can switch recovery saves off; both sides see the setting.

USD 650 (EUR 590) per project, excluding VAT, paid by card on Stripe when you accept your invitation. It includes the project licence with its AI allowance.

Each project has a modest base fee (the project licence) that includes an AI allowance; heavy AI work beyond it is added with credit packs, and Deep analysis is a seller add-on. Prices are confirmed in writing before anything is billed.

On Microsoft Azure confidential virtual machines in the EU. The approved AI model, region, retention and logging settings of each room are recorded in its Security Specification, which both sides approve before any document is read. When a room uses a model processed outside the EU (Claude models run in the United States), the specification says so. For models hosted on Microsoft Azure, Microsoft's abuse monitoring for Azure AI services applies under Azure's terms unless Microsoft has approved an exemption; the specification states which applies.