Installation

Full package :

npm install orkajs

Or install only what you need (recommended):

npm install @orka-js/core @orka-js/tools @orka-js/memory

| Package | Description | |---------|-------------| | orkajs | Full package with all features | | @orka-js/core | Types, errors, utils, Knowledge | | @orka-js/openai | OpenAI adapter | | @orka-js/anthropic | Anthropic adapter | | @orka-js/mistral | Mistral adapter | | @orka-js/ollama | Ollama adapter | | @orka-js/memory | In-memory vector store | | @orka-js/pinecone | Pinecone adapter | | @orka-js/qdrant | Qdrant adapter | | @orka-js/chroma | ChromaDB adapter | | @orka-js/agent | Agents (ReAct, HITL, Toolkits) | | @orka-js/tools | Loaders, splitters, parsers, chains | | @orka-js/cache | Memory/Redis cache | | @orka-js/resilience | Retry, fallback | | @orka-js/orchestration | Router, Consensus, Race, LoadBalancer | | @orka-js/workflow | Multi-step workflows | | @orka-js/graph | Graph workflows | | @orka-js/evaluation | Metrics, test runner | | @orka-js/observability | Tracer, hooks | | @orka-js/prompts | Prompt versioning | | @orka-js/memory-store | Conversation memory | | @orka-js/ocr | OCR & document extraction |

Quick Start

import { ReActAgent } from '@orka-js/agent';
import { OpenAIAdapter } from '@orka-js/openai';

const agent = new ReActAgent({
  llm: new OpenAIAdapter({ apiKey: process.env.OPENAI_API_KEY }),
  goal: 'Manage customer support tickets',
  tools: [
    {
      name: 'search_tickets',
      description: 'Search support tickets by status or keyword',
      execute: async ({ query }) => searchDatabase(query)
    },
    {
      name: 'send_email',
      description: 'Send email to customer',
      execute: async ({ to, subject, body }) => sendEmail(to, subject, body)
    }
  ]
});

// Agent plans, executes tools, and responds autonomously
const result = await agent.run(
  'Find all urgent tickets from last week and send follow-up emails'
);

console.log(result.output); // "Sent 12 follow-up emails"
console.log(result.steps);  // See the agent's reasoning

Features

| Feature | Description | |---------|-------------| | RAG | Semantic search & retrieval-augmented generation | | Agents | Structured agents with tools, policies, ReAct, Plan&Execute | | Workflows | Multi-step pipelines with plan, retrieve, generate, verify | | Adapters | Pluggable adapters for LLMs and vector databases | | Graph Workflows | Conditional branching with Mermaid export | | Multi-Model | Router, Consensus, Race, LoadBalancer orchestration | | Memory | Conversation memory with sliding window, buffer, summary | | Evaluation | Built-in metrics: relevance, faithfulness, hallucination | | Resilience | Retry, fallback, timeouts, circuit breaker patterns | | Caching | Memory & Redis cache for LLM and embeddings | | OCR | Extract text from images, PDFs, scanned documents (Tesseract, OpenAI Vision) | | PII Guard | Detect and redact sensitive data before LLM calls (RGPD compliant) |

Providers

LLM: OpenAI, Anthropic, Mistral, Ollama
Vector DB: Pinecone, Qdrant, Chroma, In-Memory

Documentation

📚 orkajs.com 💬 Discord

• Getting Started
• RAG & Knowledge
• Agents
• Workflows
• Best Practices

License

MIT © Orka Team