Building a Knowledge Graph Without a Graph DB: 1000x Cheaper Than GraphRAG

TheRedfather · reddit · 2026-07-22

A solo-founder shares practical learnings from building a 'company brain' knowledge base, proposing a simpler and cheaper alternative to GraphRAG.

Core Problem: Vanilla RAG handles 'find this fact' queries but fails on 'tell me everything about X' or 'list all documents mentioning Y'.

Why Not GraphRAG:

Indexing cost is extremely high; Microsoft archived their GraphRAG solution. Research shows vector-RAG indexing costs <0.1% of full GraphRAG.
Entity resolution is hard; incorrect merges propagate errors silently.
Requires maintaining a separate graph database.

Alternative Solution: Implement graph-like structures inside a regular DB (Postgres) and search index:

Use a small fixed ontology (organization, person, product, etc.).
Entity resolution follows a waterfall: alias table lookup, vector similarity, cheap LLM adjudication.
Edges are co-occurrence counts, not typed relations.
Entity summaries are lazy and cached.
Agent has four tools for different retrieval scenarios.

More from coding & agent