Skip to main content

Approving tool calls

End user & admin

For sensitive actions, an agent pauses mid-conversation and asks you to approve. This is how Ask gives you a final say before anything irreversible happens.

Prerequisites​

  • An Ask account.
  • An agent that has at least one tool that requires approval.

Why approvals exist​

Tool calls are how agents do things, and not every action is safe to do silently. Sending an email, posting to a calendar, writing to a CRM, deleting a record: these benefit from a human pause. Tool Call Approval is part of Ask's Trust & Safety capabilities; Phase 1 is live in development environments.

What an approval looks like​

  1. You ask the agent to do something, e.g. "Send a follow-up email to Sara about yesterday's meeting."

  2. The agent prepares the tool call (recipient, subject, body) and pauses. A prompt appears in the conversation showing exactly what's about to happen.

  3. You review and choose. Approving with Always allow means this tool won't prompt you again, and the stored rule lands in Personal settings → Tool permissions, where you can change it back at any time.

  4. The agent either executes the call or backs off and tells you what to do next.

Not every tool call prompts

Tools you explicitly picked in the composer, and some built-in capabilities, may run without asking. The prompt is primarily for connector tools acting on your accounts.

Manage your approval preferences​

Personal-level controls live in Personal settings » Tool permissions (profile icon, bottom-left → Personal settings): "Set when your agents are allowed to use the tools below." Tools are grouped by connector, and each tool has its own dropdown (e.g. Ask for permission):

  • Ask for permission: prompt every time. Safest default for write actions.
  • Always allow: skip the prompt. Use only for tools you trust. Choosing "always allow" from an in-chat prompt stores the same rule here.
  • Deny: block the tool entirely from your account.

Organization-level policies​

Admins can pin certain tools to always ask or block them outright org-wide. If a personal "Always allow" is overridden, you'll still see the approval prompt in conversations.

Tips & gotchas​

tip

Read the parameters in the approval prompt, not just the tool name. The agent might be sending the email to the wrong person or with the wrong content.

Heads up

Setting a write-action tool to Always allow removes a key safety net. Reserve it for tools whose worst-case behaviour is genuinely acceptable.

note

Approval prompts can interrupt long agent runs. If you regularly batch-run an agent, work with your admin to whitelist its known-safe tools at the org level.

Tools & actionsTools overviewWhere tools live and how they connect.AgentsGuardrailsAgent-level rules complement runtime approval prompts.AdminSecurity & data residencyWhere org-wide policies are set.