Configuration

Configure Orka JS defaults, adapters, memory, observability, andprompt management.

Full Configuration

Here's a complete example showing all available configuration options:

Import Methods

Standard Import

import { createOrka, OpenAIAdapter, PineconeAdapter } from 'orkajs';

Optimized Import

Recommended

import { createOrka } from '@orka-js/core';
import { OpenAIAdapter, PineconeAdapter } from '@orka-js/openai';
// Or 
//import { PineconeAdapter } from '@orka-js/pinecone';

Enables tree-shaking for minimal bundle size in production.

import { createOrka, OpenAIAdapter, PineconeAdapter } from 'orkajs';
 
const orka = createOrka({
  // Required LLM provider: 
  llm: new OpenAIAdapter({
    apiKey: process.env.OPENAI_API_KEY!,
    model: 'gpt-4o-mini',
    embeddingModel: 'text-embedding-3-small',
  }),
 
  // Required Vector database: 
  vectorDB: new PineconeAdapter({
    apiKey: process.env.PINECONE_API_KEY!,
    indexHost: 'https://your-index.svc.pinecone.io',
  }),
 
  // Optional: Default parameters
  defaults: {
    chunkSize: 1000,
    chunkOverlap: 200,
    topK: 5,
    temperature: 0.7,
    maxTokens: 1024,
  },
 
  // Optional: Memory configuration
  memory: {
    maxMessages: 50,
    strategy: 'sliding_window',
  },
 
  // Optional: Observability
  observability: {
    logLevel: 'info',
    hooks: [{
      onTraceEnd: (trace) => console.log(`Done: ${trace.totalLatencyMs}ms`),
    }],
  },
});

1. LLM Provider (Required)

The LLM adapter handles text generation and embeddings. Orka JS supports multiple providers:

OpenAI

GPT-4o, GPT-4o-mini, o1

import { OpenAIAdapter } from 'orkajs';
 
llm: new OpenAIAdapter({
  model: 'gpt-4o-mini',
embeddingModel: 'text-embedding-3-small',
})

Get API Key

Anthropic

Claude 3.5 Sonnet, Opus

import { AnthropicAdapter } from 'orkajs';
 
llm: new AnthropicAdapter({
  model: 'claude-3-5-sonnet-latest',
})

Get API Key

Mistral AI

Mistral Large, Pixtral

import { MistralAdapter } from 'orkajs';
 
llm: new MistralAdapter({
  model: 'mistral-large-latest',
})

Get API Key

OllamaLocal

Llama 3.2, Mistral, Gemma

import { OllamaAdapter } from 'orkajs';
 
llm: new OllamaAdapter({
  baseURL: 'http://localhost:11434',
model: 'llama3.2',
})

Get API Key

2. Vector Database (Required)

The vector database stores document embeddings for semantic search. Choose based on your deployment needs:

MemoryVectorAdapter

Best for: Development, testing, prototyping

import { MemoryVectorAdapter } from 'orkajs';
 
vectorDB: new MemoryVectorAdapter()

⚠️ Data is lost on restart

PineconeAdapter

Best for: Production, cloud-native, scalable

import { PineconeAdapter } from 'orkajs';
 
vectorDB: new PineconeAdapter({
  apiKey: process.env.PINECONE_API_KEY!,
  indexHost: 'https://your-index.svc.pinecone.io',
})

→ Get Started with Pinecone

QdrantAdapter

Best for: Self-hosted, on-premise, Docker

import { QdrantAdapter } from 'orkajs';
 
vectorDB: new QdrantAdapter({
  url: 'http://localhost:6333',
  apiKey: process.env.QDRANT_API_KEY,
})

→ Qdrant Documentation

ChromaAdapter

Best for: Local development, embedded use

import { ChromaAdapter } from 'orkajs';
 
vectorDB: new ChromaAdapter({
  url: 'http://localhost:8000',
})

→ Chroma Documentation

3. Default Parameters (Optional)

Set default values for chunking, retrieval, and generation. These can be overridden per-call.

4. Memory Configuration (Optional)

Configure conversation memory to maintain context across multiple interactions. The sliding_window strategy keeps the most recent messages.

5. Observability (Optional)

Monitor your application with logging and trace hooks. Use hooks to integrate with external monitoring tools like DataDog, New Relic, or custom analytics.

Configuration Reference

Property	Type	Default	Description
`chunkSize`	number	1000	Characters per chunk
`chunkOverlap`	number	200	Overlap between chunks
`topK`	number	5	Results for RAG queries
`temperature`	number	0.7	LLM creativity (0 to 1)
`maxTokens`	number	1024	Max tokens per response

💡 Pro Tip

All default values can be overridden per-call. For example, orka.ask({ topK: 10 }) overrides the default topK for that specific query.