Requires a BIG AI contract. If BIG AI is not active on your account, this feature will not appear.
BIG AI can answer from your own documents and web pages, not just loan data. Admins curate what it is allowed to read, and every answer that uses the knowledge base shows its sources.
Where to find it
Admin Portal → AI Admin → Company Documents.
Two source types
Company documents — upload and organize the files the assistant may read, with folder-level organization and PDF previews in the media library. Default folders can be renamed and deleted like any other. Larger documents are accepted than before, so a full guideline set no longer has to be split up to be indexed.
URL sources — point BIG AI at a live web page instead of maintaining a copy as an uploaded file. Sources refresh automatically each night, and citations link to the live page with an as-of date. Tables and collapsed sections on a page are kept intact, so matrix and overlay pages index the way they read, and an answer drawn from a table can quote the relevant rows back to you.
RSS and Atom feeds are accepted. Many rate-commentary and news sites build their pages in the browser, leaving nothing to read from the page itself; the feed carries the same content in an indexable form.
Domain approval
The standard sources Big POS ships are approved out of the box. A custom domain goes through an approval step before it can be fetched. A source on an unapproved domain sits as pending rather than failing.
Asking across every investor or program
A question covering all investors or all programs now searches every lender folder, and more searches are run per answer. If a folder could not be reached, the answer names it, so you can tell an incomplete answer from a complete one.
Per-prompt grants
Each prompt gets access only to the folders you grant it, so a guideline prompt and a product prompt can read different material.
Note that every approved, active source is indexed whether or not a prompt has been granted it. Grants control which prompts may read a source, not whether it is fetched. If you approved sources you never intended a prompt to use, review them.
Re-index is available on any approved, active source, so you can confirm a URL works before wiring it to a prompt.
A prompt that a task relies on for document verification is protected and cannot be deactivated while that dependency exists, so the tasks built on it keep working. See articles 16 and 17.
Isolation
Your knowledge base is isolated to your account. Nothing is shared or blended across clients.