Skip to main content
Agents can use ShingleAI tools — things like “read contact,” “send email,” “create task.” You control what each agent is allowed to do on a per-tool basis.
Tool permissions tab

​
The three permission levels

Each tool on each agent is set to one of three values:
LevelIconBehavior
AllowGreen checkAgent calls the tool silently — no interruption
AskBlue questionAgent pauses and requests your approval before calling
DenyRed XTool is hidden from the agent entirely
The “Ask” approval appears inline in the chat with a countdown timer and a risk-level badge. You can expand the card to see the exact tool input before approving.

​
Finding the permissions UI

  1. Open the agent’s detail page.
  2. Go to the Tools tab.
  3. Tools are grouped by category (for example, Contacts, Messages, Tasks). Expand a category to see its tools, each with a three-way button.

​
How defaults work

When you create an agent, it inherits the regular-user baseline permissions for the organization. Tools considered sensitive default to Ask; lower-risk tools default to Allow. Your overrides are stored sparsely — only the tools where you deviate from the default are persisted. If the default shifts later (for example, we promote a new tool from risky to routine), agents inherit the new default unless you had explicitly overridden that tool.

​
Risk levels

Tool approval cards display a risk badge (low / medium / high) to help you decide quickly. Use your judgment: a low-risk tool misused is rarely damaging, a high-risk tool might send email from your domain or spend money.
Changing permissions takes effect on the next tool call by the agent. A tool call already mid-flight when you change the setting will complete under the old rules.

​
Next steps

Chat with an agent

See approvals in action

Agent conversations

Review past approvals and denials