Reflection

Enable agents to review and improve their own work

Reflection lets agents review their output and self-improve.

Quick Start

1. Create Self-Reviewing Agent

   use praisonai::Agent;
   
   // Build reflection loop into instructions
   let agent = Agent::new()
       .name("Writer")
       .instructions("Create high-quality content by:
       1. Writing an initial draft
       2. Reviewing your draft critically
       3. Making improvements based on your review
       4. Providing the polished final version")
       .build()?;
   
   let response = agent.chat("Write about AI").await?;
   // Agent writes, self-reviews, and outputs improved version

2. Two-Agent Reflection Loop

   use praisonai::Agent;
   
   // Writer agent
   let writer = Agent::new()
       .name("Writer")
       .instructions("Write clear, engaging content")
       .build()?;
   
   // Reviewer agent for reflection
   let reviewer = Agent::new()
       .name("Reviewer")
       .instructions("Review the content. List 3 specific improvements needed.")
       .build()?;
   
   // Reflection loop
   let mut content = writer.chat("Write about AI trends").await?;
   
   for _ in 0..2 {  // Max 2 improvement rounds
       let feedback = reviewer.chat(&format!("Review:\n{}", content)).await?;
       content = writer.chat(&format!("Improve based on feedback:\n{}\n\nOriginal:\n{}", feedback, content)).await?;
   }

How It Works

Configuration

Best Practices

Related