GraphRAG.jsPure TypeScript Graph RAG

Quick Example ​

import { createGraph } from "@graphrag-js/core";
import { similarityGraph } from "@graphrag-js/similarity";
import { memoryStorage } from "@graphrag-js/memory";
import { openai } from "@ai-sdk/openai";

const graph = createGraph({
  model: openai("gpt-4o-mini"),
  embedding: openai.embedding("text-embedding-3-small"),
  provider: similarityGraph(),
  storage: memoryStorage(),
});

await graph.insert("Your documents...");
const { text } = await graph.query("Your question?");

How It Works ​

1. Vector Search (Baseline) ​

All algorithms start with vector similarity search to find relevant chunks:

Query → Embedding → Vector Search → Top K chunks

2. Graph Enhancement (Different Strategies) ​

Each algorithm builds and uses graph structures differently to enhance these results:

3. Context Generation ​

Graph-enhanced context is richer than pure vector search:

• Connected chunks: Include neighbors in the graph
• Relationship context: Add descriptions of how entities relate
• Community insights: Include higher-level summaries
• Multi-hop paths: Follow chains of related information

Why Graph-Enhanced RAG? ​

Pure vector search retrieves similar chunks. Graph structures add:

• Better context: Include connected information, not just similar text
• Relationship awareness: Understand how entities and concepts connect
• Global reasoning: Access community-level summaries and patterns
• Multi-hop queries: Follow relationship chains to find indirect connections

GraphRAG.js brings multiple graph enhancement strategies to TypeScript with a unified API.

Algorithm Comparison ​

Choose based on your use case:

// Simple: BFS expansion from similar chunks
provider: similarityGraph({ similarityThreshold: 0.7 })

// Balanced: Dual-level entity + relation vectors
provider: lightrag({ entityTypes: ["person", "org", "location"] })

// Deep analysis: Community detection and reports
provider: microsoftGraph({ graphClusterAlgorithm: "leiden" })

// Fast & cheap: PageRank-based expansion
provider: fastGraph({ pagerank: { damping: 0.85 } })

// Multi-hop: Fact-centric graph traversal
provider: awsGraph({ semantic: { beamWidth: 5 } })