You do not switch products when you switch places. The question you ask on the vault screen is the same one your AI client asks through the agent channel, and the same one your system asks through the API. What changes is the door; what answers is always the same engine reading the same vault.

Three doors, one conversation

On the vault screen, in the AI client you already use, or straight from your system

On screen, you talk to the agent next to the documents it is reading: you ask, you get the answer with the source, you click the source and land on the slide. In your AI client, the vault becomes one more thing that client knows how to query, without leaving where you already work. And through the programming channel, your team automates from outside exactly what a person does by talking.

the screenthe agent next to the documents
one vaultthe same engine, the same reading
your client, your systemthrough the agent channel or the API
Three doors, and behind all three the same vault with the same reading.

In all three, the answer arrives the same way: with the address of where it came from, and with a refusal when the subject is not in your documents.

The agent channel is an open standard

Authentication and discovery by public standards, not by a convention of our own

The channel through which an AI client talks to your vault is not a house invention. It is a resource server in the web’s standard authorization design: whoever arrives without a credential gets 401, and the document that says how to authenticate is published for any client to read on its own. A convention of our own was dropped on purpose, because it would require every integration to learn our way before asking the first question.

401The agent channel is liveThe MCP endpoint answers 401 without a credential, which is the correct behavior for an OAuth resource server, and discovery at .well-known answers 200. That is the machine talking, not us.
A 401 here is proof the door exists and is locked, not that it fell.

One substrate, several readers

A new channel does not invent a new surface

An open API and an agent channel are not plumbing: they are what lets other interfaces get built on the same substrate, without us in the room. A third-party agent, a panel for your team, a workflow integration. The product gets to be those things because the doors exist, not because we foresaw each one.

And what keeps the doors from turning into different products is a rule that holds as a contract: the set of operations exposed to the agent is a subset of what the API exposes. Nothing exists in one channel without existing in the other, so a new capability does not appear on one side and vanish on the other. The whole set, and what it refuses, is in capabilities.

The door changes; the owner of the vault does not

The conversation stays with you, wherever you speak from

From any of the three doors, the vault is yours: the same isolation between organizations, the same metering, the same record of who did what. The conversation stays with you, and nothing the agent answers is built with another company’s data, because it has no source beyond your vault.

The boundary of this sentence gets stated here and not in a footnote: the text that goes to the model follows the provider’s retention terms, and the agent’s own first line already says so. What we do with your data, and what you can audit on your own, is in privacy.

These are the doors. What passes through them, operation by operation, is in capabilities, and the reason the file is never touched by the model is in the conversation is with the AI.