Criteria - PraisonAI

Agents can validate their work against criteria before returning results.

Quick Start

Simple Usage

import { Agent } from 'praisonai';

const agent = new Agent({
  instructions: 'Write blog posts',
  criteria: 'Must be at least 500 words and include 3 key points'
});

const result = await agent.chat('Write about AI trends');
// Agent validates output meets criteria before returning

With Configuration

const agent = new Agent({
  criteria: [
    'Contains at least 500 words',
    'Includes an introduction',
    'Has actionable takeaways'
  ]
});

User Interaction Flow

Configuration Levels

// Level 1: String - Simple criteria
const agent = new Agent({
  criteria: 'Must be professional and concise'
});

// Level 2: Array - Multiple criteria
const agent = new Agent({
  criteria: [
    'At least 300 words',
    'Includes examples',
    'No jargon'
  ]
});

// Level 3: Dict - With options
const agent = new Agent({
  criteria: {
    rules: ['professional', 'concise'],
    maxRetries: 3,
    strict: true
  }
});

Criteria Options

Option	Description
`rules`	List of validation rules
`maxRetries`	Attempts before failing
`strict`	Fail on any criteria miss

API Reference

AgentConfig

Agent configuration including criteria

Best Practices

Be specific

“At least 500 words” is better than “long enough”.

Keep criteria measurable

Criteria the agent can verify work best.

Set retry limits

Prevent infinite loops with maxRetries.

Reflection

Self-improvement

Output

Output formats

​Quick Start

​User Interaction Flow

​Configuration Levels

​Criteria Options

​API Reference