Type @ and pick an agent
Researcher, Quizmaster, Charts, or any agent your team has defined. Teammates are in the same menu.
Agents
Agents that work where you write.
No copying prompts into a chat window and answers back out. Ask in the page, and the page gets the answer.
Write what you need
Press Enter. The request becomes a block on the page, and everyone sees the agent working.
The answer replaces it
Formatted, with sources, in the page. If it fails you can retry, and you can stop it any time.
In the page
Pick an agent from the @ menu.
The same menu lists your agents and your teammates, in separate sections. Mention a teammate and they get an email and an inbox notification, and the editor offers to share the page if they can't open it yet.
The result
Answers with sources, written into the page.
Here the Researcher answered a question for tomorrow's starter, with a table and linked sources. The page header shows it as a co-author.
Your agents
Define the agents your team needs.
Owners and admins give each agent a name, a colour, instructions, an optional model and the tools it may use. An agent for marking, one for market research, one for turning notes into tickets.
Global agents
Set them up once, every workspace has them.
Instance admins create agents that every workspace can call, such as a Reporter that writes status updates from daily logs, or a Planner that drafts today's plan. Each workspace can switch any of them off.
Found one that's close? Customize copies it into your workspace with your own instructions and your choice of model, and it replaces the global one there.
Agent Mode
Or work in a chat, with approvals.
For bigger jobs, chat with the agent. It proposes each change to your workspace and waits for your go-ahead, or you can let a chat run on auto-approve.
In control
Powerful, and on a short leash
Agents are useful because they can act. These are the rules that keep that safe.
🔐
Your permissions, no more
An agent sees and changes only what the person asking can, and only in one workspace.
✋
It asks first
In Agent Mode, changes wait for your approval. Publishing, making things public and deleting always ask.
💰
A budget you can see
Each workspace has a monthly AI budget. A single request can't spend more than its share, and the chat shows what's used.
🧭
A scope check on every message
Requests are checked first, so the agent stays on workspace work. Admins set the policy.
🪪
Credit where it's due
Pages show which agents wrote on them, and for whom, so nothing appears without a trail.
🔌
Tools you choose
Per agent: read the workspace, search the web, edit other pages, or use trusted connectors.
Outside the app
Your workspace, as a tool for any assistant
Connect claude.ai, ChatGPT or any MCP client to a workspace and your pages, collections and forms become something that assistant can read and write, with the same permissions as you. Sign in once with OAuth and pick the workspace, or use a token for scripts.
- One URL per workspace
- Read and write pages, collections and forms
- Permissions follow the person who connected
- Revoke access from the workspace any time
See an agent work on your material
Start a 14-day free trial, no card needed. Prefer a walkthrough first? We'll show you a setup that fits your work.