Knowledge and access

How many documents a single research run may read, 200 by default, adjustable from 1 to 1000. A scope over the limit is declined before it runs, with the limit named, so each run stays fast, focused, and a deliberate choice. The ceiling reflects how much one run can read live and well, not a storage limit: Research reads each document in scope where it lives and never copies it.

The two Knowledge tools answer differently. Research is non-deterministic: it reads and reasons over your documents, weighing and interpreting, so a run’s cost scales with how many it reads, which is what the cap bounds. Structured Query is the deterministic sibling: it counts over fields you prepared ahead of time, so the same question over the same data returns the same answer every time, without re-reading every document. The cap governs the read-and-reason work, where bounding scope keeps each run honest and fast.

Who can ask over a folder follows who can see it. A folder set up for the whole organization is visible to everyone; a folder scoped to specific departments is visible only to people in those departments. Visibility is enforced at the database, so a folder a person can’t see never appears in Research or Structured Query for them, and never leaks through a question.

This area also holds the curated folder collections: the named sets of drive folders an administrator manages, each scoped to the whole organization or to specific departments. Here you create a collection, add or remove the drive folders it draws from, set who can see it, sync its document inventory, and prepare it for exact answers. Folders people pick directly in Research and Structured Query are managed there; these curated collections are the way to scope a set of folders to specific departments. Creating and editing them is super admin only; other administrators see them read-only.

The research cap is super admin only; org admins see it read-only.