Praison AI

AutoRag Agent

Automatic retrieval-augmented generation with smart query detection

Automatically retrieve relevant context based on query analysis with AutoRagAgent.

Diagram

AutoRag Flow

Needs Context

No Context

📝 Query

🔍 Analyze

📚 Retrieve

🤖 Direct

🧠 Generate

📤 Response

Quick Start

  1. Create AutoRagAgent
    typescript
    import { AutoRagAgent, RetrievalPolicy } from 'praisonai';

const autoRag = new AutoRagAgent({
  retrievalPolicy: RetrievalPolicy.AUTO,
  topK: 5,
  rerank: true
});
  2. Check if Retrieval Needed
    typescript
    // Queries that trigger retrieval
autoRag.shouldRetrieve("What is machine learning?");  // true
autoRag.shouldRetrieve("Explain the document");       // true
autoRag.shouldRetrieve("Find information about AI");  // true

// Queries that don't trigger retrieval
autoRag.shouldRetrieve("Hi");                         // false
autoRag.shouldRetrieve("Thanks");                     // false

Retrieval Policies

Diagram

Choose Your Policy

Smart detection

Every query

Never

When to retrieve?

🔮 AUTO

📚 ALWAYS

🚫 NEVER

typescript
import { RetrievalPolicy } from 'praisonai';

RetrievalPolicy.AUTO    // Smart detection based on query
RetrievalPolicy.ALWAYS  // Always retrieve context
RetrievalPolicy.NEVER   // Never retrieve context

Configuration

typescript
import { AutoRagAgent, AutoRagAgentConfig, RetrievalPolicy } from 'praisonai';

const config: AutoRagAgentConfig = {
  retrievalPolicy: RetrievalPolicy.AUTO,
  topK: 5,                    // Number of results to retrieve
  hybrid: false,              // Use hybrid search
  rerank: true,               // Rerank results
  includeCitations: true,     // Include citations in response
  citationsMode: 'append',    // 'append' | 'hidden' | 'none'
  maxContextTokens: 4000,     // Max tokens for context
  autoMinLength: 10           // Min query length for auto mode
};

const autoRag = new AutoRagAgent(config);

Configuration Options

OptionTypeDefaultDescription
retrievalPolicyRetrievalPolicyAUTOWhen to retrieve context
topKnumber5Number of results to retrieve
hybridbooleanfalseUse hybrid (keyword + semantic) search
rerankbooleanfalseRerank results for relevance
includeCitationsbooleantrueInclude source citations
citationsModestring'append'How to show citations
maxContextTokensnumber4000Max context window
autoMinLengthnumber10Min query length for AUTO

Auto Keywords

Queries containing these keywords trigger retrieval in AUTO mode:

typescript
import { DEFAULT_AUTO_KEYWORDS } from 'praisonai';

// Default keywords that trigger retrieval
const keywords = [
  'what', 'how', 'why', 'when', 'where', 'who', 'which',
  'explain', 'describe', 'summarize', 'find', 'search',
  'tell me', 'show me', 'according to', 'based on',
  'cite', 'source', 'reference', 'document', 'paper'
];

Custom Keywords

typescript
const autoRag = new AutoRagAgent({
  retrievalPolicy: RetrievalPolicy.AUTO,
  autoKeywords: new Set([
    'what', 'how', 'explain',
    'my custom keyword',
    'another trigger'
  ])
});

Methods

shouldRetrieve

Check if a query should trigger retrieval:

typescript
const autoRag = new AutoRagAgent();

// Returns true for questions
autoRag.shouldRetrieve("What is the capital of France?");  // true

// Returns true for search queries
autoRag.shouldRetrieve("Find documents about AI");         // true

// Returns false for short queries
autoRag.shouldRetrieve("Hi");                              // false

// Returns false for non-question statements
autoRag.shouldRetrieve("Thanks for your help");            // false

getConfig

Get the current configuration:

typescript
const config = autoRag.getConfig();
console.log(config.topK);           // 5
console.log(config.retrievalPolicy); // 'auto'

Common Patterns

With Knowledge Base

typescript
import { AutoRagAgent, RetrievalPolicy, Knowledge } from 'praisonai';

const knowledge = new Knowledge();
await knowledge.add('./docs', { userId: 'u1' });

const autoRag = new AutoRagAgent({
  retrievalPolicy: RetrievalPolicy.AUTO,
  topK: 5,
  rerank: true
});

async function answer(query: string) {
  if (autoRag.shouldRetrieve(query)) {
    const found = await knowledge.search(query, { userId: 'u1', limit: 5 });
    return generateWithContext(query, found.results);
  }
  return generateDirect(query);
}

With Agent

typescript
import { Agent, AutoRagAgent, RetrievalPolicy } from 'praisonai';

const autoRag = new AutoRagAgent({
  retrievalPolicy: RetrievalPolicy.AUTO
});

const agent = new Agent({
  name: "RAG Assistant",
  instructions: "Answer questions using retrieved context"
});

async function chat(query: string) {
  const needsContext = autoRag.shouldRetrieve(query);

  if (needsContext) {
    // Add context to the query
    const context = await retrieveContext(query);
    return agent.start(`Context: ${context}\n\nQuestion: ${query}`);
  }

  return agent.start(query);
}

API Reference

AutoRagAgent

AutoRagAgent class reference

 AutoRagAgentConfig

Configuration options

Best Practices

Use AUTO for general chatbots

AUTO mode balances retrieval costs with response quality for most use cases.

Use ALWAYS for document Q&A

When every query relates to documents, use ALWAYS to ensure context is included.

Enable reranking for accuracy

Reranking improves result relevance at a small latency cost.

Tune autoMinLength

Increase autoMinLength to avoid retrieval for very short queries.

Related

RAG Agent

Full RAG implementation

 Knowledge Base

Build knowledge bases