Notion is where a team writes down what it knows
Notion is where a team writes down what it knows, and it is also where that knowledge ages without anyone noticing. The process page has three versions, two of them are wrong, and nothing says which one holds.
An AI plugged into that workspace inherits the problem. It will answer with the version it found, unaware that the version is stale, and it will do so with confidence. That is why connecting a Notion workspace is not enough, and why what matters is knowing which pages are authoritative, and for whom.
Nothing comes in until you have shared the page
The Notion connector works the opposite way round from what people expect. It does not roam your workspace looking for things it could read. It sees strictly the pages a human has shared with it, through the same sharing menu you would use for a colleague.
The starting scope is therefore empty, and it grows by successive decisions. A team that wants to begin with its client knowledge base shares that base, and nothing else in the workspace is even listed.
What you share, and nothing beyond it
The Notion connector's scope is the one you give it, and it starts out empty.
Removing a page from the share is enough to take it out of scope: it drops out of the listing and your Kastel marks it as gone at the source. Content already taken in nonetheless stays reachable there by those who already had access, until an administrator decides to erase it.
Notion does not tell us who is allowed to read a page
Here is the limit, written down rather than worked around. The programming interface Notion offers integrations does not expose the list of people allowed to read a page. So we cannot inherit Notion's permissions the way we inherit Google Drive's, where every file carries its own.
The signal we use instead is a person-type property on the page, the one teams use to name an owner or a recipient. When it is filled in, the page is filed under the department of those people. When it is not, the page is filed nowhere and waits for an administrator.
We do not treat authors and last editors as rights holders, and that is a choice. Having edited a page does not mean deciding its audience, and conflating the two would be the quietest way to widen access without anyone deciding to.
The practical consequence shows up on the very first connection. A loosely structured Notion workspace produces a lot of pages on hold, and someone then has to decide, page by page or in batches. That is real work, and it is the price of a scope you can prove to an auditor rather than assume.
A database row is a page
In Notion, a database row is a page in its own right, with properties and content. The connector treats rows as such, which makes readable what teams actually keep in their databases, the meeting notes, the client records, the decisions.
Text alone comes in. Images and files attached to a page are not downloaded, even when the page is. A mockup dropped into a page therefore stays in Notion, and the text around it is what your Kastel keeps.
What this connector does not do
It does not guess a page's audience. With no permissions exposed by Notion, a workspace where the person property is never filled in ends up with every page on hold, and there is real filing work waiting for you. A connector that filed those pages on your behalf would be pleasanter on the first try and wrong at the first audit.
Connecting is an administrator's gesture. The access key is created in your Notion workspace settings, at workspace level, not by each person for themselves. That is the opposite of the mail connector, where everyone connects their own mailbox.
Notion publishes no change feed, so every pass lists the shared pages again. Unchanged pages cost nothing to take in a second time, but a very large workspace takes longer than a source able to say what moved. And reading is the only verb: Kastel writes, edits and archives no page.
See what Kastel does with a workspace.
The product page shows how scattered knowledge becomes structured, governed context that the AI of your choice can query, without you having to hire a data team.
See the product
