Approval - PraisonAI

Agents can request human approval before taking sensitive actions.

Quick Start

Simple Usage

import { Agent } from 'praisonai';

const agent = new Agent({
  instructions: 'You can execute commands',
  approval: true  // Requires approval for all tool calls
});

// Agent will pause and ask for approval
await agent.chat('Delete the temp files');
// → "Approve action: delete_files? [y/n]"

With Configuration

const agent = new Agent({
  approval: {
    autoApprove: ['read_*', 'search_*'],  // Safe patterns
    autoDeny: ['delete_*', 'rm_*']        // Dangerous patterns
  }
});

User Interaction Flow

Configuration Levels

// Level 1: Bool - Require approval for everything
const agent = new Agent({
  approval: true
});

// Level 2: Array - Require for specific tools
const agent = new Agent({
  approval: ['delete', 'execute', 'send']
});

// Level 3: Dict - Auto-approve/deny patterns
const agent = new Agent({
  approval: {
    autoApprove: ['read_*', 'list_*'],
    autoDeny: ['delete_*'],
    timeout: 60000
  }
});

// Level 4: Instance - Custom handler
import { ApprovalManager } from 'praisonai';

const approver = new ApprovalManager({
  handler: async (request) => {
    return await askUserViaSlack(request);
  }
});

Approval Options

Option	Description
`autoApprove`	Patterns to auto-approve
`autoDeny`	Patterns to auto-deny
`timeout`	Time to wait for response (ms)
`handler`	Custom approval function

API Reference

ApprovalCallbackConfig

Complete configuration options

Best Practices

Auto-approve safe actions

Reading data is usually safe to auto-approve.

Auto-deny dangerous patterns

Block delete, execute, and similar risky operations.

Set reasonable timeouts

Default to 60 seconds; adjust based on your workflow.

Guardrails

Input/output validation

Security

Security features

​Quick Start

​User Interaction Flow

​Configuration Levels

​Approval Options

​API Reference