Ask before risky tools

Gate destructive tools with toolApproval and render the request.

AI SDK docsai-sdk.dev
Tool execution approval lets you require user confirmation before a server-side tool runs.
AI SDK UI · Chatbot Tool UsageRead docs ↗

From ai-sdk and the Tool Calling docs.

"Delete my latest file"

Press Send, then approve or deny in the Good pane.

Runs on its ownBad
    tools: { deleteFile }
    Needs approvalGood
      toolApproval: { deleteFile: 'user-approval' }
      // app/api/chat/route.ts
const result = streamText({
  model: 'anthropic/claude-sonnet-5',
  messages: await convertToModelMessages(messages),
  tools: { deleteFile },
  toolApproval: { deleteFile: 'user-approval' },
});

// app/page.tsx
const { messages, addToolApprovalResponse } = useChat({
  sendAutomaticallyWhen: lastAssistantMessageIsCompleteWithApprovalResponses,
});

switch (part.state) {
  case 'approval-requested':
    return (
      <div key={part.toolCallId}>
        Delete {part.input.path}?
        <button onClick={() => addToolApprovalResponse({ id: part.approval.id, approved: true })}>Approve</button>
        <button onClick={() => addToolApprovalResponse({ id: part.approval.id, approved: false })}>Deny</button>
      </div>
    );
  case 'output-denied':
    return <p key={part.toolCallId}>Kept {part.input.path}.</p>;
}

      Why it matters, pros and cons

      Why it matters

      The model picks the input. A wrong guess on a delete, a payment or an email can't be undone.

      Pros

      • The person sees the exact input before it runs.
      • The tool still runs on the server, with its secrets.
      • A denial goes back to the model, which can answer it.

      Cons

      • An extra click on every call: gate only what can't be undone.
      • Approval needs a second model call to resume.
      • needsApproval on the tool is deprecated in v7.