AI in your Workspace

AI in PageSpace isn't bolted on — it's a behaviour the whole product shares. Any AI Chat page can be @mentioned from any document, channel, sheet cell, or other chat; agents can consult other agents; and when an agent acts, it acts under your permissions, using real workspace tools.

What you can do

  • @mention an AI Chat page from a document, channel, sheet cell, or another AI Chat to pull that agent into the thread.
  • Use the global assistant in the right-hand sidebar to work across every drive you can see.
  • Ask one agent to consult another by name — the called agent runs under its own prompt and tools, then returns a single response.
  • Pick your model in Settings > AI from one catalogue spanning every vendor — your account-level pick sets the default, and any individual AI Chat page can override it.
  • Restrict any agent to a specific toolset by editing its AI Chat page's allow-list.
  • Toggle Read-only mode on any AI Chat page to guarantee it cannot create, edit, or delete anything.
  • Toggle Web search on any AI Chat page to let that agent look things up outside your workspace.
  • Undo what an agent did on any turn — preview the edits that assistant message caused, then revert them as a group.

How it works

Models. PageSpace gives you one catalogue of models from many vendors — OpenAI, Anthropic, Google, xAI, and more — all in a single picker. PageSpace holds the credentials, so you never paste an API key or manage a provider account. Each call draws from your plan's monthly AI-credit allowance based on the model's real cost; free accounts pick from a curated set of efficient models, while paid plans unlock the full catalogue across every vendor, plus top-up credits whenever you need more.

Permissions. When an agent acts, it acts as you. It can only see pages you can see and only change pages you can change. Share a page with a teammate and their agents can now see it too, under their account. Revoke access and the agents lose access immediately — there is no AI back-door.

Tools. Every AI Chat page carries an allow-list of tools it may use. If the list is empty, the agent can chat but cannot act. The read-only toggle strips every write tool (create, edit, delete, send) on top of that list; the web-search toggle adds or removes the one web-lookup tool.

@mentions. Any AI Chat page can be @mentioned from a document, channel, sheet cell, or another chat. Mentioning an agent inside a channel pulls it into that thread as a participant — it reads the context around the mention and replies there.

Agent-to-agent. One agent can consult another by name. The called agent runs under its own system prompt and tools, but still with the calling user's permissions, and returns a single response. Chains are depth-capped so agents can't spiral into each other forever.

Related

  • AI Chat — the page type the agent lives in, and its per-page configuration.
  • Channels — where @-mentioning an agent pulls it into a live thread.
  • Task Lists — assigning work to an agent so it picks it up.
  • Sharing & Permissions — the rules an agent inherits when it acts.
  • Search — what the agent's search tools can reach.
  • MCP Integration — connecting external AI clients like Claude Desktop and Cursor.

Search

Search docs, blog posts, and more.