Spaces

Spaces let users organize chats by context and isolate knowledge between different working domains. A single agent often serves multiple roles: work projects, personal tasks, creative writing. Spaces solve two problems at once: they organize the chat list, and they create knowledge boundaries.

How it works

A space belongs to an agent and has a name, symbol, instructions, and a privacy setting. Private spaces seal their chats and knowledge from the rest of the agent's context. Non-private spaces are organisational only. The instructions are injected into the per-turn hint block whenever the user is chatting within that space, giving the agent persistent context about the current domain without the user needing to repeat it.

Organization

Spaces group chats by context. A user might have a "Work" space, a "Personal" space, and a "Project X" space. The chat list can be filtered by space.

Knowledge isolation

Knowledge created within a private space is only accessible within that space. Knowledge created anywhere else is accessible everywhere. For details on how knowledge scoping works, see Knowledge - Scoping.

Creating a space

Spaces can be created from within a normal conversation or through a dedicated setup conversation. Either way, the agent interviews the user about the space's purpose, picks a name and symbol, sets instructions, and creates the space via a tool call.