An open standard file format for agent-readable data rooms and CIMs.
Agents are already in the data room. Adding a VDR.md file is how you put your best foot forward for that AI audience.
VDR.md is simply a markdown formatted file (or optionally multiple files) to include in your virtual data room. It sits alongside all of your usual content: CIM, financials, product demos, etc. However, unlike those traditional documents, the VDR.md file is intended for a different audience: AI agents.
Bankers already draft with AI. On the buy side, associates summarize with it, and diligence teams point agents at the data room to find what matters faster.
What those agents get is hundreds of PDFs, spreadsheets and scans, plus a CIM written to persuade a reader in print, where logos and charts carry the numbers and the sentences are cut to fit a slide.
They read it unevenly. They drop charts, misread a figure through OCR or invent words that were never on the page, lose the thread between one file and the next, and spend a lot of tokens doing it.
The fix is to put something in the room that is written for them.
Other conventions already work this way. robots.txt tells crawlers the rules of a site. security.txt tells researchers where to send a report. llms.txt and AGENTS.md give language models and coding agents one file, in a known location, written for the machine reading it. VDR.md does the same for the room where a company is sold.
The CIM and the general documents in a data room are built for people: layout, a chart, a photograph, numbers that live inside a picture. VDR.md restates the same facts as text an agent can read without guessing.
---
Spec: https://vdrmd.org/v/2026-09-draft
Title: Pied Piper, Inc. - Finances
Version: v1.3
Last-Modified: 2026-09-15
Author: Donald "Jared" Dunn <jared@piedpiper.com>
---
# Finances - Pied Piper, Inc.
All figures are in US dollars (USD). The fiscal
year ends December 31.
### Figure 1.2: Monthly ARR, January-June 2026 (USD)
Source: @1.1 sheet "ARR Bridge"
```yaml
arr_monthly:
- month: 2026-06
arr: 48612300
- month: 2026-05
arr: 47280150
```
---
Spec: https://vdrmd.org/v/2026-09-draft
Title: Pied Piper, Inc. - Data Room Overview
---
# Pied Piper, Inc.
Pied Piper, Inc. is a Delaware corporation that develops and sells
encryption software for enterprise data at rest and in transit. All
figures are in US dollars and the fiscal year ends December 31.
## Contents
- Financial statements: FY2025 P&L and balance sheet.
@"1 - Finances/1.1 - Statements.xlsx"
- Org chart: headcount by team as of 2026-06-30.
@"2 - People/2.1 - Org Chart.pdf"
- Product overview: what it does and who buys it.
@"3 - Product/3.1 - Overview.pdf"
The file below is the whole thing. Two header lines, the company name, a paragraph saying what the business does, and a short list pointing at documents your room already holds. There is nothing to install, nothing to compile and no account to make. It is a text file.
VDR.md, spelled exactly that way. In Notepad, change “Save as type” to “All Files” first, or Windows will save it as VDR.md.txt and it will not count. On a Mac, if TextEdit offers to add .txt, choose “Use .md”.That is a conforming file, and everything else on this page is something you can add later. In practice most VDR.md files are written by an agent or produced by a tool rather than typed out by hand. The format stays plain text so that you can open one, check what it says and change it yourself.
Ten short sections, each one read against the same sample data room.
Or jump to the full spec →01
At its core, VDR.md is simply a text document in markdown formatting that includes much of the content you might otherwise stuff into a pretty deck. It sits alongside the standard data room material, and it doesn't replace it.
The root VDR.md file is the companion to the CIM: same scope, more detail. It names the company, says what it does, and carries two required sections: Contents, a map of the room, and Definitions, where the conventions live.
Remember, this is for AI, not a human reader, so pack in all of the important detailed content that was left on the cutting room floor because it didn't fit into the PowerPoint slide.
02
The fields at the top are written like email and HTTP headers, because those parsers exist in every language and hand-editing is forgiving. Only Spec and Title are required. Spec is the URI of the specification version the file follows, so an agent that has never seen the format can fetch it and learn how to read what it is holding. These are recommended:
Last-Modified and Version, so staleness is visibleAuthor, repeated once per person, and Contact for who to askAccess-Tier, the data room permission level the file was written forExpires, after which the file shouldn't be relied onGoverned-By, the confidentiality terms that apply to itMetadata stays metadata. Currency, fiscal year and what each metric means belong in Definitions. Instructions is the one field that may address the reader, and only about how to consume the material.
03
A printed CIM is limited by what a person will read. VDR.md isn't, so it keeps the detail the printed version cuts: every product line, every legal entity, every metric with its history.
Use absolute dates (2026-06-30, not "last quarter"), full entity names, and a unit on every number.
04
The root file's Contents section maps the room and links to a deeper VDR.md in each folder. An agent answering a question about revenue follows one link and skips the rest.
References use an @ and a path in quotes, like @"1 - Finances/VDR.md". The path is relative to the file doing the linking.
05
The same @ syntax references the source documents themselves: the CIM, the P&L, the penetration test report. A Source: line names the file and the page, sheet or slide where relevant.
Where the room numbers its documents, the index number alone is enough: @4.2 resolves to the penetration test report and survives a rename. The source documents remain authoritative. VDR.md tells you where to check.
06
This is 1 - Finances/VDR.md. An agent may be pointed straight at it and never see the root file, so it restates its own context: who the company is, the currency, the fiscal year, and which figures are audited.
Repetition is encouraged. It costs a few lines and saves the agent from guessing.
07
Raw data goes in fenced YAML, JSON or CSV blocks, such as revenue history, churn, feature adoption or code metrics.
Each block gets a numbered Figure title so it can be referenced from anywhere in the room, and a Source: line that names the spreadsheet and sheet it came from.
Ordinary Markdown tables are allowed anywhere and read well for something short and flat. Anything worth referencing later, and anything wide or nested, belongs in a fenced block instead: a table flattens structure, loses types, and carries no figure id to point at.
08
ARR, net revenue retention and adjusted EBITDA mean slightly different things at different companies, and every room has its own vocabulary. VDR.md settles them with ordinary CommonMark link definitions at the foot of the file.
They render as nothing, so they cost the human reader no space. The destination is where the number comes from, and the title is the definition, which a renderer shows on hover wherever prose writes [ARR].
No new syntax and no section to remember. Define anything a reader might otherwise guess at, and point at the file that settles it.
09
VDR.md follows the data room's permissions. Each file is written for one access tier and is complete at that tier, so a buyer's agent never gets a summary of documents the buyer can't open.
Sellers decide whether to say that restricted material exists. Pied Piper says so in its technology file: source code is available in a clean room at tier 2, and nothing from it is summarized here.
10
A standard for agent-readable deal material only works if buyers trust it. VDR.md contains no instructions to the agents reading it, only declarative facts with sources.
A VDR.md can never be more favorable than the data room behind it. Any risk disclosed in the room is disclosed in VDR.md, including a high-severity pen test finding.
VDR.md tells the data room's story in more detail, never a different one.
How conformance gets checked, how the draft is being shaped, and who is shaping it.
Every rule in the specification is numbered, so a linter can cite the one you broke. It resolves each reference, confirms that every figure carries a title, a source line and exactly one fenced block, and separates what is invalid from what is merely unwise.
Run it over the room before you send it: npx vdrmd:lint path/to/VDR.md
The linter is running against this sample room. See the linter and the rest of the tools →
We're inviting bankers, sellers, buyers and data room providers to use it, challenge it and help shape it. Open-source agent skills and the linter are coming next.
The first data room to adopt it will make the next one easier.
VDR.md was started by Crosslake, which spends its days in data rooms helping investors understand the technology behind the businesses they buy. It's developed in the open with these members.
Organizations
Have a thought to share? Want to support the cause? Request to join the consortium.