The difference between a Knowledge base that gets referenced constantly and one nobody remembers exists.
Most Knowledge bases die the same way: someone dumps in fifty documents on day one, nobody curates it afterward, and within a month it's a pile nobody trusts enough to reference. A smaller, well-maintained Knowledge base beats a large, stale one every time — AI responses citing it are only as good as what's actually in there.
Add the handful of documents your team references most — a positioning doc, a policy, a pricing sheet — not everything you have. Let it grow from actual use (agents promoting session artifacts, people adding sources as questions come up) rather than a one-time upload.
| If it's... | Use |
|---|---|
| A living document that changes often | Google Drive connection — stays in sync automatically |
| A static reference (a signed contract, a finished report) | Upload directly |
| Something your team already produced in a session | Promote the artifact — no re-uploading |
| A fact or policy that doesn't exist as a file anywhere | Paste it in as text |
See Choosing what to connect for a deeper framework on this.
Set a recurring reminder — quarterly is reasonable — to review what's in Knowledge and remove anything outdated. A workspace whose Knowledge base includes a pricing sheet from two plans ago is worse than one with no pricing sheet at all, since AI responses will confidently cite the wrong number.
Sessions opt in to workspace Knowledge — it's not automatically visible everywhere. This is useful, not just a restriction: it means a sensitive Knowledge base (say, HR policy documents) doesn't leak into unrelated sessions just because it exists in the same workspace.