AI Assistant to Search Our Google Drive Documents by Meaning (OpenAI + Vector Search)

We are a small consulting company (about 10 people). All our documents live in one shared Google Drive folder - proposals, contracts, internal guides. About 1,500 files, and the folder grows every week. The problem: nobody can find anything. We remember what a document was about, not what it was called, so people search for ten minutes and then ask a colleague. Half the questions in our team chat are "where is the file about X". What we need - a simple internal tool: - We ask a question in plain language, the tool answers from our documents and shows which documents the answer came from, with a link to open the file in Drive. - If the answer is not in our documents, it must say so. We would rather get "not found" than a made-up answer. - When a new file is added to the folder, it appears in the search by itself (within minutes is fine). When a file is deleted, it disappears from the search. - Documents are text only (PDF, Word, Google Docs), English only. No images, no scans. - We want to use it from Slack too, so people can ask right in our team chat instead of opening another tool. To keep the first version simple - this is a proof of concept: - No login and no user roles. The tool is for our team only, we will not expose it publicly. - No saved chat history. You can ask follow-up questions in one session, but it all lives in the browser: refresh the page and it is gone. Nothing is stored on the server. - One page: a search box and the answer. Looks do not matter, this is an internal tool. - Our own OpenAI key. Our documents are confidential and must not be used to train anything. - Runs on our small VPS, or recommend something cheap. - Short README at handover so our IT person can keep it running. If this first version proves itself, we will keep extending it from there. When applying: have you built an assistant over company documents before? How do you make sure it does not invent answers?