Modular Plugin Architecture
5 plugin types with JSON Schema contracts and 7 provider connectors (OpenAI, Anthropic, Cohere, Ollama, and more). Swap any component without rewriting your pipeline.
Smart Chunking & Retrieval
5 chunking strategies, hybrid vector + BM25 search, and reciprocal rank fusion for highly relevant context retrieval across large document sets.
RAG Evaluation
Faithfulness, relevance, context precision/recall, and groundedness metrics to measure and improve RAG pipeline quality.
Quick Start
Build a complete RAG pipeline in minutes. Install the package, configure your plugins, and start querying your documents with AI-powered retrieval.
npm install @devilsdev/rag-pipeline-utils
import {
createRagPipeline,
OpenAIConnector,
MemoryRetriever,
} from '@devilsdev/rag-pipeline-utils';
const pipeline = createRagPipeline({
retriever: new MemoryRetriever(),
llm: new OpenAIConnector({ apiKey: process.env.OPENAI_API_KEY }),
});
const result = await pipeline.run({
query: 'What is the vacation policy?',
options: { citations: true, evaluate: true },
});Three Paths, One Pipeline
Start with retrieval. Add evaluation when quality matters. Add guardrails when you ship. Each primitive is optional — pick what your team actually needs.
Answer questions with citations
Return grounded answers where every sentence links back to the source chunk. Per-sentence citation tracking and groundedness scoring built in.
const result = await pipeline.run({
query: "Which policy covers remote work?",
options: { citations: true },
});
// result.citations maps each answer
// sentence to its source documentsAssistants with faithfulness guarantees
Score every answer on faithfulness, relevance, context precision, and recall. Alert when quality drops below your production threshold.
const result = await pipeline.run({
query: userQuestion,
options: { evaluate: true },
});
// result.evaluation.scores
// = { faithfulness, relevance, ... }Pipelines with production guardrails
Three-layer defense: prompt-injection detection pre-retrieval, relevance thresholds during retrieval, and PII + groundedness checks post-generation.
const safePipeline = new GuardrailsPipeline(
createRagPipeline({ ... }),
{
preRetrieval: { enableInjectionDetection: true },
retrieval: { minRelevanceScore: 0.6 },
postGeneration:{ enablePIIDetection: true },
}
);Production-Ready Features
Modular Plugin Architecture
5 plugin types with JSON Schema contracts and 7 provider connectors (OpenAI, Anthropic, Cohere, Ollama, and more).
Learn more →Smart Chunking & Retrieval
5 chunking strategies, hybrid vector + BM25 search, and reciprocal rank fusion for highly relevant context retrieval.
Learn more →RAG Evaluation
Faithfulness, relevance, context precision/recall, and groundedness metrics to measure and improve pipeline quality.
Learn more →Citation & Grounding
Source attribution, hallucination detection, and groundedness scoring to ensure trustworthy AI-generated answers.
Learn more →Agentic RAG & GraphRAG
Query planning, iterative retrieval, self-critique loops, and knowledge graph traversal for complex multi-hop questions.
Learn more →3-Layer Guardrails
Prompt injection detection, ACL filtering, and PII detection to protect your pipeline at input, retrieval, and output stages.
Learn more →Streaming & Connectors
SSE/WebSocket streaming plus built-in OpenAI, Anthropic, Cohere, and Ollama connectors for flexible deployment.
Learn more →Cost & Debugging
Token tracking, budget enforcement, execution tracing, and bottleneck detection for full pipeline observability.
Learn more →Enterprise Security
JWT replay protection, plugin sandboxing, audit logging, and multi-tenancy support for production deployments.
Learn more →Ready to Build Production RAG Systems?
Join developers building enterprise-grade RAG pipelines with modular architecture, comprehensive security, and production observability.