Research on building a self-hosted knowledge graph for intelligent search across research documentation.

Documents

DocumentDescription
Architecture and CostsFull analysis of FalkorDB + GraphRAG-SDK approach
Implementation ComparisonComparison of @research/graph vs alternatives (LightRAG, GraphRAG, LlamaIndex)
FalkorDB Architecture & StorageDeep dive into FalkorDB’s Redis dependency and storage flexibility
Graph Database ComparisonFalkorDB vs DuckDB vs SQLite: when to use each
DuckDB Hybrid SearchCombining VSS + DuckPGQ for vector + graph queries
Naming: LatticeBrainstorming and selection of marketable project name

Summary

The Goal

  1. Ask questions about research topics
  2. Search knowledge graph for existing answers
  3. Research online if not found, update docs
  4. Auto-extract entities as documents change

Solution Stack

ComponentChoiceCost
Graph DatabaseFalkorDB (Docker)Free
SDKGraphRAG-SDKFree
Entity ExtractionGPT-4o-mini or Haiku0.85

Key Decisions

  1. Vector vs Graph RAG: Start with Graph RAG for future-proofing
  2. Self-hosted: FalkorDB over Zep Cloud (cost + control)
  3. SDK over custom: GraphRAG-SDK saves days of development
  4. Model choice: GPT-4o-mini ($0.12) for initial testing

Quick Start

Terminal window
# 1. Start FalkorDB
docker run -p 6379:6379 -p 3000:3000 -it --rm falkordb/falkordb


# 2. Install SDK
pip install graphrag_sdk


# 3. Ingest docs (~$0.12)
python ingest.py