Set up an AI agent with an approval mode
Create an AI agent in AI Studio, choose how much it may do on its own — Assist, Semi or Autonomous — and approve the actions it pauses for.
An AI agent works in your workspace with the tools its skills give it. Its autonomy level decides which actions it may take on its own and which it pauses for, waiting until a person approves.
How the autonomy levels work
APROS sorts every action an agent can take into one of these kinds, and each level allows a different set without asking:
| The agent wants to… | Assist | Semi | Autonomous |
|---|---|---|---|
| Read data | runs | runs | runs |
| Create or change data in APROS | asks | runs | runs |
| Change custom fields or record types | asks | asks | runs |
| Send something outside APROS — an email, a document to a client, a public share link | asks | asks | runs |
| Take another action that is hard to undo, short of deleting | asks | asks | runs |
| Delete data | asks | asks | asks |
A new agent starts at Assist. Deleting always needs a person's approval, at every level.
At Autonomous, the agent sends real email and sends documents to clients without asking. Start at Assist or Semi, watch what the agent asks for, and raise the level only when you trust its work.
Tools from MCP servers you connect count as sending data outside APROS.
Before you begin
| Permission | What it allows |
|---|---|
VIEW_AI_STUDIO | Open AI Studio. |
CREATE_AI_STUDIO, UPDATE_AI_STUDIO | Create agents, and change their settings and skills. |
MANAGE_AI_AGENT_SECURITY | Change the Autonomy level, delegation, task work and schedules. As shipped, only the Administrator role has it. |
APPROVE_AI_EXECUTION | Approve or deny paused actions in runs other people started. Administrators, Directors and Department Heads have it. |
AI Studio belongs to the AI module. The AI Brain stays available when the module is off.
1. Create the agent
-
In the sidebar, click AI Studio, stay on the Agents tab and click New Agent. APROS creates an agent called New Agent right away and opens its Configure tab.
-
Under Basic Information, set the Name and Description, and pick an emoji or an Agent Color.
-
Under Personality & Instructions, describe what the agent does, what it knows and the rules it follows.
-
Under Model Configuration, keep the default Model or choose another one.
-
Under Chat, keep Available in Chat on if people should talk to the agent in Chat. Under Memory, Long-term Memory lets the agent remember facts across conversations.
-
Click Save Configuration.
2. Give it skills
An agent can use only the tools its skills give it.
-
Open the agent's Skills tab and click Add Skill.
-
Select the skills it needs and click Add Selected. The change is saved at once.
The Tools tab lists the agent's Effective tools and which skill each comes from.
3. Choose the autonomy level
MANAGE_AI_AGENT_SECURITY-
On the Configure tab, scroll to Coordination & Delegation.
-
In Autonomy level, choose Assist, Semi or Autonomous.
-
Click Save Configuration.
The level applies wherever the agent works: in Chat, on tasks assigned to it and on its schedules. The AI Brain uses its own switch — see below. When the AI Brain or another agent hands work to this agent, the more permissive of the two levels applies.
4. Put the agent to work
- Talk to it: click Run agent at the top of the agent's page. APROS opens a conversation with the agent in Chat. The button is unavailable while Available in Chat is off.
- Give it a task: in a task's assignee field, switch to AI agent and pick the agent. For the agent to work on tasks, turn on Task work on the agent's Work tab.
- Run it on a schedule: on the Work tab, under Schedules, click + New schedule.
5. Approve or deny what it asks
When the agent reaches an action its level doesn't allow, it stops and waits. You can decide in three places:
- In the conversation: a Confirm action card shows the action and its details. Click Allow or Deny.
- In AI Studio → Approvals: the tab shows a count of waiting actions. The Pending list says which agent wants to do what, with the details, who started the run and where it came from. Click Approve or Deny. Archive lists past decisions.
- From a notification: APROS notifies you when an agent needs approval; the notification opens the approval.
After Approve (or Allow), the agent continues. After Deny, the action isn't performed and the run stops; a task the agent was working on stays blocked. Once you approve an action, the agent doesn't ask again for exactly the same action with the same details in the same run.
Who can decide: the person who started the run, and anyone with APPROVE_AI_EXECUTION. People with that permission see every waiting action in the workspace; others see only their own.
A waiting action expires after 24 hours. The run is then cancelled, and the expired request appears in Archive.
The AI Brain's Approval and Auto
When you work in the AI Brain, the switch in the message box decides, for that conversation, instead of the agent's own level:
- Approval — "I ask for permission before important actions." It works like Assist. This is the default.
- Auto — the AI Brain works on its own, like Autonomous: it asks only before deleting data.
Your choice is remembered in this browser.