What is an agent-readable room?
An agent-readable room is a shared buyer and seller workspace that an AI assistant can connect to directly and read as structured data, under a credential a person issued and can withdraw. Instead of a person copying status into a chat window, their own assistant reads the room's record: what was agreed, what was confirmed, and what is waiting on whom.
Three things separate it from an interface that merely exists. The assistant's access is no wider than what its owner may see. What it reads is marked as data written by people, never as instructions to follow. And the acts only a person should make, such as confirming an outcome or sending an invitation, stay a person's click.
ValueRoom's rooms work this way today, over the Model Context Protocol (MCP), with two separate doors. A buyer can connect their own assistant to their room with a credential they create inside it, once the vendor has opened that door for the room. It is read-only, and the buyer can withdraw it at any time. A member of the vendor's team can connect their own assistant with a credential from their settings, and the admin decides what that assistant may propose.
The reason
Why it exists
People on both sides of a deal now work through assistants.
If the assistant cannot read the room, the person becomes the copy and paste layer between the two, and the assistant reasons from a summary that is already out of date. A room an assistant can read keeps it working from the same record the people see.
It also has to be safe to allow. A buyer will not connect an assistant to a room that reports back to the seller, and an admin will not let an assistant act where it could reach a customer unasked. The limits are what make the door usable.
Often confused with
How it differs from the adjacent instruments
A chatbot inside a product
Belongs to the vendor's product and lives on its screen. An agent-readable room is reached by the person's own assistant, from wherever that person works.
A public API
Built for developers to write integrations against. This door is built for an assistant acting for one person, with that person's view and no more.
A browser agent clicking through the page
Reads the screen and guesses at structure. An agent-readable room hands over the record itself, in a published shape.
A CRM integration
Syncs fields between two systems on a schedule. This is one person's assistant reading, and where allowed proposing, on that person's behalf.
Questions
Frequently asked
Can an AI assistant confirm a milestone or sign for someone?
No. There is no tool for approving, confirming, co-signing, inviting or sending. Those stay a person's click. A vendor's assistant can propose routine moves, such as drafting a room or adding a note, and the admin sets whether each kind of move waits for a person.
Does the seller see what a buyer's assistant reads?
No. Reading the room through the buyer's door records nothing toward the vendor. The buyer's own card shows when their credential was last used, and only the buyer sees it.
Is there anonymous access?
No. Every connection uses a credential a person created, and the public site hands out none. A buyer's credential comes from inside their room, and a vendor seat's from that person's settings.
What can a buyer's assistant read?
What the buyer's own view of the room contains: the plan and what has been confirmed, the co-signed value statement, the people on the customer's side, and what the vendor published for them. It cannot read the vendor's scoring, risk view or strategy.
Enforced, not just promised. The guarantees behind these terms (append-only records, the records-nothing buyer lane, the buyer-side firewall) are enforced in our system and checked by automated tests. Read the live trust protocol
Try it from your customer's side
A real room with the attestation layer, no credit card. Your own accounts follow a 30-minute scope meeting.