Vatan Knowledge
Your documents, under the same rules as everything else.
Upload what your agents need to know. Grant a person a knowledge base and their agents inherit exactly that, no more. Every retrieval says which passages it returned and which document each came from, on the same key and the same bill as your model access.
One header, no code change
Add x-vatan-knowledge to a chat completion you already make with the OpenAI SDK and the answer is grounded in your documents. No second base URL, no second credential, no rewrite.
A key is never wider than its person
Bind a key to an employee and its reach is the intersection of what it was scoped to and what they currently hold. Revoke the person and the key stops on the next request, with nothing to redeploy.
You can see what the model saw
Which passages were retrieved, at what score, from which document, and which survived into the prompt. The difference between a corpus and a black box that occasionally lies.
Documents are scanned on the way in
A document is retrieved into prompt after prompt, so a planted instruction inside one is a far better payload than a hostile chat message. Every file is scanned at upload and every passage reaches the model marked as data.
What this is not
It is not a better embedding model, and we would rather say so than imply otherwise. The retrieval here is ordinary, and any competent team builds the same pipeline in a fortnight. What is hard to assemble, and what this is for, is one identity, one policy and one log across the models your agents call, the tools they use and the documents they read. If you already have that, you do not need this.