Reflection

Agents that improve their own work

Agents can review and improve their own responses before returning.

reflection is accepted on a Task for Python-SDK parity but is not yet implemented in the TypeScript SDK — passing it emits a [praisonai] … not yet honoured notice and has no effect. This feature is Python-only for now.

Quick Start

1. Simple Usage

   import { Agent } from 'praisonai';
   
   const agent = new Agent({
     instructions: 'You write professional content',
     reflection: true
   });
   
   await agent.chat('Write a business proposal');
   // Agent writes, reviews, and improves before returning
   
2. With Configuration

   const agent = new Agent({
     reflection: {
       rounds: 2,  // Review twice
       prompt: 'Check for clarity and professionalism'
     }
   });
   

---

User Interaction Flow

---

Configuration Levels

// Level 1: Bool - Enable with defaults
const agent = new Agent({
  reflection: true
});

// Level 2: Dict - With options
const agent = new Agent({
  reflection: {
    rounds: 2,
    prompt: 'Review for errors and clarity'
  }
});

// Level 3: Instance - Full control
const agent = new Agent({
  reflection: {
    rounds: 3,
    prompt: 'Focus on tone and accuracy',
    stopWhen: (output) => output.quality > 0.9
  }
});

---

How It Works

1. Generate: Agent creates initial response
2. Review: Agent critiques its own work
3. Improve: Agent fixes identified issues
4. Repeat: Process continues for specified rounds

---

API Reference

ReflectionConfig

Complete configuration options

---

Best Practices

Use 1-2 rounds

More rounds don’t always mean better results.

For quality-critical tasks

Enable for proposals, articles, and important content.

Combine with criteria

Use reflection with criteria for validated quality.

---

Related

Criteria

Define success conditions

Agent

Create agents