“A secure data room is a feature you rent from someone else.” Flip that assumption and a whole category of software gets cheaper. The pitch here is a self-hosted, encrypted deal-room service — rooms that expire on schedule, enforce who-sees-what down to the page, bill per room, and answer questions about their own contents — assembled from capabilities that already ship on VBWD rather than bought from four separate vendors. The room is the product. Sovereignty over the documents is the whole differentiator.

Encrypted document storage, metered billing, expiring access, and an AI that reads the room are usually four procurement decisions. The VBWD original — a self-hosted, source-available full-stack SaaS SDK — reframes them as four toggles against one agnostic core, wired through the office, subscription, booking, RBAC, and mcp plugins. Here is why that matters.

The room is an encrypted object you control

The office plugin is a self-hosted, encrypted document suite, and in M&A, legal discovery, or a fundraise, the sensitive material is the point. The moment cap tables and board minutes live in a vendor’s multi-tenant cloud, your customer is trusting that vendor’s security model and jurisdiction. Run office on infrastructure you control and that trust problem disappears. As the original analysis puts it:

For a diligence process, “our documents never left our own tenancy” is not a nice-to-have; it is frequently the thing that closes the deal.

RBAC and booking: who, and when

A data room without granular access control is just a shared folder. VBWD’s core ships RBAC and multi-tenancy, so sell-side counsel sees everything, a bidding party sees only its released tranche, and an outside auditor gets a read-only slice with no export — roles defined once and enforced by the core.

The non-obvious move is time. The booking plugin exists for appointments, seats, and slots, and a deal room is a room with a start and an end. The source makes the case cleanly:

Model each engagement as a booking: access opens when the diligence window opens and closes when it closes.

A “48-hour final review” becomes a 48-hour booking. Identity decides what; the time window decides when; the two compose, and every open, extend, and close fans out through the core’s native signed webhooks to your logging and the client’s own monitoring.

Bill per room, then let the room answer itself

Deal-room pricing is naturally per-engagement. The subscription plugin handles tarif plans, add-ons, and recurring billing, so a room is a monthly line while the deal is open, extras are add-ons, and billing stops at close. Every charge is an invoice line item against the one shared interface the whole platform uses — a single point of truth instead of four subsystems keeping separate ledgers. And because the payment core is provider-agnostic, you settle through a regional card processor, an invoice-and-wire arrangement, or non-custodial crypto to your own wallet.

The piece that feels like science fiction is mcp, which turns platform capabilities into agent-callable tools. Point it at the room and a permitted user asks, “Which contracts have a change-of-control clause?” — grounded in the actual documents, and bounded by the same access model:

The room answers questions, but it answers them within RBAC, not around it.

One honest caveat: an LLM over your documents is an accelerant, not an oracle. Treat its answers as a first pass a human confirms, and write that into your contract with the client.

What you still build

VBWD is a substrate. It gives you encrypted storage, roles, time-boxed access, per-room billing, and an agent seam that already respect one another. You still bring the onboarding flow, the disclosure playbook, the branding, and you still run the ops of self-hosting — backups, key management, uptime, audits. What it removes is the eighteen months of undifferentiated plumbing between “we have documents” and “we have a billable, expiring, self-answering deal room.” Read the full analysis on VBWD, or start from the public SDK.

Written for Red Robot with AI assistance and human editing. Adapted from an original analysis published on VBWD.