@flowrag/pipeline

Indexing and querying pipelines for FlowRAG with dual retrieval (vector + graph).

Installation

npm install @flowrag/pipeline

Usage

import { createFlowRAG } from '@flowrag/pipeline';
import { createLocalStorage } from '@flowrag/presets';
import { defineSchema } from '@flowrag/core';

const schema = defineSchema({
  entityTypes: ['SERVICE', 'DATABASE'],
  relationTypes: ['USES', 'PRODUCES'],
});

const rag = createFlowRAG({ schema, ...createLocalStorage('./data') });

// Index documents
await rag.index('./content');

// Search with dual retrieval
const results = await rag.search('how does auth work');

// Trace data flow
const flow = await rag.traceDataFlow('auth-service', 'downstream');

// Statistics
const stats = await rag.stats();

Query Modes

• hybrid — vector + graph combined (default)
• local — focus on specific entities found in query
• global — high-level concepts enriched with graph keywords
• naive — vector search only, no knowledge graph

Indexing Pipeline

Files → Scanner → Chunker → Extractor (LLM) → Embedder → Storage

• Incremental indexing (SHA-256 content hashing)
• LLM extraction caching
• Configurable concurrency

License

MIT