Using Space Agents
What agents are
AI workers with a job. What they can do, the three kinds, and why they can never do more than you can.
An agent is an AI worker with a defined job. Where the copilot in chat is a general helper, an agent has a name, a role and instructions, such as “You prepare the weekly sales summary and flag anything unusual”. You hand it a task, and it works through it using the tools and knowledge you've allowed.
What an agent can do
- Take on tasks you or a colleague give it, and report back.
- Use tools and your Brain: search knowledge, read tables, send messages through connected apps.
- Ask for help: pause and ask you a question, ask another agent, or request a review.
- Run on its own from a schedule or a trigger, such as “every Monday, prepare the report”.
- Learn from feedback, if you leave that switched on. See Agents that learn.
Three kinds of agent
| Kind | Belongs to | Who edits it | Who can give it tasks |
|---|---|---|---|
| Personal agent | You | You | You |
| Team agent | A group | The group's admins (and Org Admins) | Every member of the group |
| Front Desk | You | You | Other people, but only after you approve each request |
- A personal agent is your own assistant for your own work.
- A team agent is shared: a “Support triage” agent that everyone on the support team can use.
- Your Front Desk answers requests from colleagues on your behalf. See Front Desk. Everyone gets one automatically. It can be paused but not deleted.
The safety rule: never more than the person
An agent acts as a person: its owner, or for a team agent, whoever gave it the task. It can never do more than that person could do themselves.
- If you can't see a file, neither can your agent.
- If your organization switched off a tool for you, your agent can't use it.
- If you leave a group, you can't use that group's team agent any more.
This is enforced by the system, not by trusting what the agent is told. Access is re-checked every time the agent runs, and again every time it uses a tool.
Choosing what an agent can use
On the agent's Resources tab, you choose one of two settings:
| Setting | The agent can use |
|---|---|
| Everything I can use (default) | Whatever you can, and that updates as your access changes. |
| Only what I pick | Only the knowledge bases, skills, playbooks, tools, tables and so on you tick, and only while you still have access. |
Choosing Only what I pick is the better choice for agents with a narrow job, or ones handling sensitive data.
Mentioning an agent
In chat, type @ and the agent's handle to involve it: “@ada, prepare the weekly plan.” Handles are short names, unique per owner or group.
Where to see your agents
- Agents in the sidebar lists them, with each agent's profile, resources, permissions, links, tasks and improvements.
- Control Room shows them live while they work. See Control Room.
Next steps
- Agent orchestration and communication: how agents are arranged and talk to each other
- Create an agent
- Tasks and human checks
- Agents working together