Collections and ownership

What it is

Two different ideas share this page, and keeping them apart avoids most of the confusion people have about where their knowledge lives.

Collections are folders. They group sources on one chatbot so a long Knowledge table stays navigable. Purely organisational — a collection does not change what the chatbot can answer, and it does not give you extra capacity.

Ownership is the rule about where a source actually lives: every source belongs to one chatbot, in one workspace. There is no shared library, no cross-chatbot linking, and no way to attach the same source to a second chatbot. If two chatbots need the same document, you add it twice.

Why there is no shared library

It looks like a missing feature until you consider what a chatbot is. Each one has its own knowledge cap, its own training runs, and its own answers. A shared source would have to be counted against two caps, retrained for two chatbots, and removed from one without breaking the other. The simpler rule — one source, one chatbot — makes it unambiguous which material can produce which answer, and that turns out to matter far more than the convenience of a shared pool.

The practical cost is duplication when you genuinely need the same content twice. The practical benefit is that "why did this chatbot say that?" always has an answer you can find on one table.

When you would use it

• The Knowledge table has grown past the point where you can find anything.
• You want to filter down to one product line, one language, or one department while you work.
• A teammate says they cannot see a file you uploaded.
• You uploaded to the wrong chatbot and want to know how to fix it.

Where to find it

The collection picker lives in the Add drawer and in the edit drawer for an existing source: Dashboard → Chatbots → your chatbot → Knowledge → Add.

The Knowledge table itself can filter by collection.

Steps

1. Open Dashboard → Chatbots → your chatbot → Knowledge → Add.
2. Find the collection picker among the shared fields at the top of the drawer.
3. Choose an existing collection, or create a new one from the option at the foot of the menu. Creating one opens a small dialog on top of the drawer.
4. Give it a name that will still make sense to a colleague in six months — "Refunds and returns" rather than "misc 2".
5. Save the source into that collection.
6. Back on the Knowledge table, use the collection filter to narrow the list.
7. To move an existing source, open its row, change the collection, and save. That is a metadata change and does not retrain anything.

What you will see

The collection control sits inside the Add and edit drawers, not on the table toolbar. Creating a collection opens a stacked dialog above the drawer; if it appears to do nothing, look for a validation message — an empty name is the usual cause.

The table's collection filter narrows what you see. It does not narrow what the chatbot searches. A filtered table is a view for you, not a restriction on answers.

How to organise, in practice

Collections earn their keep once you are past a dozen or so sources. Some groupings that work:

• By subject — Pricing, Shipping, Returns, Technical. Best for most support chatbots.
• By product line — one collection per product, when your catalogue is the main axis.
• By source — Website crawl, Uploaded PDFs, Written by us. Useful when you want to prune crawled noise later without hunting for it.
• By language — when one chatbot serves several languages and you need to audit each set separately.

Two habits worth adopting early: put crawled pages in their own collection so you can clear them in one pass when you re-crawl, and keep a small "Corrections" collection for the short text sources you write to fix specific wrong answers. Both make later cleanup much less painful.

When a teammate cannot see your file

Almost always one of three things, in this order:

1. They are in the wrong workspace. Knowledge belongs to a workspace, and a person who belongs to several has to switch into the right one. See Switch accounts.
2. They are on the wrong chatbot. Sources do not appear on any chatbot but the one they were added to.
3. Their role does not include knowledge access, or their membership is scoped to a subset of chatbots that does not include this one. See Roles and permissions.

What it is almost never: a collection. Collections do not hide anything from anyone.

Limits and plan notes

• Collections do not add capacity. Your knowledge cap is per chatbot regardless of how many collections you create. See Knowledge limits.
• A Viewer can see sources and collections. Creating, editing, and removing them needs a role that permits managing knowledge.
• Sources never move between chatbots, and never between workspaces. Deleting a workspace's owner account takes that workspace's knowledge with it — see Delete your account.
• The same document on two chatbots counts against both caps separately.

Common problems

A teammate uploaded into the wrong chatbot.

Open that chatbot's Knowledge, remove the row, and add it again on the correct chatbot. There is no move. If the file is large, get the original from them rather than downloading and re-uploading a copy of a copy.

Create category does nothing.

The dialog is showing a validation error, most often an empty or duplicate name. Fix the name and confirm.

I filtered by collection and the chatbot stopped answering.

It did not. The filter changes your view of the table only. Clear the filter to see everything again.

I want the same knowledge on five chatbots.

Add it to each one. Where the shared content is substantial, consider whether those five really need to be separate chatbots — one chatbot with clear instructions often serves better than five near-identical ones, and it costs one knowledge budget instead of five.

Renaming a collection broke my links.

It does not. Collection names are labels; nothing outside the dashboard refers to them.