Space search lets the AI Agent use keyword and semantic search to find relevant content in your Space. It is a search setting, not a switch that disables every way the Agent can read content. The sections below explain search results, conversation context, and the current page separately.
Searchable content
When Space search is on, the Agent can use search tools to discover relevant content in the Space. Keyword and semantic search have different supported object types and searchable fields; a search result is a way to find content, not a guarantee that the Agent has read its complete contents.
Chat and message instances placed on a whiteboard are not searched as separate items like regular cards. The Agent can read their content when it reads the whiteboard that contains them.
To use a particular conversation, add its whiteboard with the + button and name the conversation in your request. For information you expect to find repeatedly, save the key conclusions in a regular card.
What the Agent reads
The Agent decides which searches and reads are needed for your request. It can search, inspect results, read selected objects, and repeat these steps to gather more information.
A semantic search returns up to 20 results in one call, subject to its output budget. This is not a 20-card limit for an entire question. The Agent can make additional searches and read more objects during the same task. Searching a Space with 1,000 cards does not automatically send the full contents of all 1,000 cards to the model.
Tool activity in the chat helps you see searches and reads the Agent performs. However, the absence of a particular searched or viewed label is not proof that the model received no other context: attachments, content already in the conversation, and current-page information may also be available.
Current-page context
When you send a message, the Agent may receive information about the page you are viewing. Depending on the page, this can include:
The object type, ID, and title of the current card or tab.
The current page number of a PDF.
A whiteboard’s visible area and references to selected or focused objects.
The title and URL of a web tab.
The date of the journal or the selected tag.
These details help the Agent understand references such as “this PDF” or “the selected cards.” They are not the same as the full text of every object on screen. The Agent can use reading tools when it needs the underlying content.
Turn off Space search
Turning Space search off disables the keyword and semantic Space-search tools. It does not remove your messages, attached context, or information already supplied in the conversation.
It also does not disable every reading or listing tool. For example, the Agent can still use tools to read a known card, whiteboard, or PDF. Do not interpret Space search being off as a guarantee that only manually attached objects can be accessed.
If you want the Agent to focus on a particular source, attach it and explicitly state the scope of the task. That is a way to guide the work, not an access-permission boundary.
Sensitive content
Space search does not provide a per-card or per-whiteboard exclusion list. A request such as “do not read this card” is an instruction to the Agent, not an enforced permission rule.
To keep a task focused, start a new conversation, attach only the sources you intend to use, and describe the allowed scope. You can turn Space search off to prevent keyword and semantic searches, but this does not create a restricted workspace or revoke other reading capabilities.
If you need a guarantee that particular content cannot be accessed by AI, do not rely on the Space search toggle or a prompt as that guarantee. Contact support to confirm the available controls before using AI with that requirement.
Work with filtered cards
Card Library filters cannot currently be attached to an AI conversation as a single context source. To work with the filtered result set:
Select the relevant cards in Card Library and add them to a dedicated research whiteboard.
Start a new AI Agent conversation and add that whiteboard with the + button.
State that the task should focus on the cards in that whiteboard. Turn Space search off if you do not want keyword or semantic searches across the Space.
For hundreds of cards, split the research into smaller subtopics and review each result. This approach helps the Agent focus; it is not a technical restriction on all content it can access. There is no fixed 20-card reading limit for the entire task.
If you have questions about your workflow or data-access requirements, please contact our support team.