Knowledge & the AI
How notebook pages flow into answers: the “Include in every answer” switch, automatic retrieval via vector search, citations and permissions.
The actual purpose of the notebook: your knowledge should reach the AI's answers. There are two paths for this – an explicit one and an automatic one. Both respect the permissions of the levels.
Path 1: “Include in every answer”
On the right of every page’s header sits the Include in every answer switch. Its hint reads: “This page’s content is automatically added as context to every AI answer — ideal for glossaries, policies and company facts.” When enabled, the page content flows into every answer – no matter what the question is about. In the page tree you recognise such pages by the green Always on badge.
This is ideal for things that should always apply:
Company: tone of voice, spellings, product names, compliance principles.
Team/project: context every answer in that team or project environment must know.
Personal: “about me”, preferred language, formatting preferences.
Which of these pages are used depends on the chat: in a private chat, those of the company, your teams and your personal ones; in a project chat only those of the company and that project; in a team chat only those of the company and that team. Kasimir combines them in a fixed order – company → team → project → personal – with the most specific taking precedence: your personal preference beats the general company default.
Use sparingly! Everything included in every answer consumes context space on every single question. If such a page gets very long, a yellow exclamation mark appears at the switch; its tooltip then shows characters and approximate tokens and advises “very long, please shorten”. If there is too much in total, individual pages may be dropped for space reasons. Rule of thumb: short, dense cheat sheets instead of extensive manuals.
Path 2: Automatic retrieval
For all other pages you do not need to do anything. Every knowledge page is automatically vectorised (turned into a searchable form) when saved. If you ask a question in chat that matches the content of one of your pages, Kasimir finds the relevant passages on its own and uses them for the answer – without you attaching or mentioning the page.
How the division of labour works in practice:
“Include in every answer” | Automatic retrieval | |
|---|---|---|
When active | On every question | Only when the question matches the page content |
Suited for | Short principles, preferences, style rules | Domain knowledge, how-tos, reference material |
Effort | Enable the switch, keep it short | None – just write |
New and changed pages are usually findable via automatic retrieval after about a minute – Kasimir adds them to the search index in a regular pass. Pages with “Include in every answer” apply immediately.
Sources: answers stay traceable
If an answer uses content from the notebook, the pages involved appear under Sources below the answer – with the page title, a violet dot and the label Knowledge page. This way you always see what the AI relied on, and can correct the page if needed: the next answer then uses the updated state.
Permissions apply to the AI as well
The AI sees exactly what you are allowed to see – nothing more:
Team pages only flow in for team members – in their private chats and in the team’s chat.
Project pages apply only in that project’s chats.
Personal pages are only known to the AI in your private chats – not in project or team chats.
Company pages are available to all members of the workspace in every chat.
So there is no detour through which knowledge from one level reaches unauthorised people – neither by reading nor via AI answers.