Agent needs to access thousands of documents as a Knowledge source. Where should I host them?

You need to find every document containing a keyword and produce a spreadsheet, so the main requirement is reliable indexed search rather than conversational answers. One Microsoft-supported option is Azure Blob Storage for the files, with Azure AI Search indexing their extracted text and metadata. Start with representative documents, including scanned PDFs. Configure a blob indexer and retain the document name and storage path as retrievable fields. Add OCR where the content is an image rather than selectable text. Test keyword matching carefully: analyzers can change how punctuation, word forms, and phrases match. Query the index, retrieve every results page, and have your application write the document names and matched terms into the spreadsheet. Preserve source permissions in the retrieval design. Your agent can orchestrate this workflow, but its generated answer should not be treated as proof that every matching document was found.