PageIndex vs Graphify: Structured Retrieval Tools Compared

PageIndex, open-sourced by VectifyAI in April 2025 under MIT, builds each document into a hierarchical tree index and lets an LLM reason about which nodes to read at query time — no vector database, no text chunking.

Graphify, maintained by Graphify Labs under Apache-2.0, turns a codebase plus its docs, SQL schemas, configs and PDFs into a queryable knowledge graph: code goes through local deterministic tree-sitter parsing, and every edge is tagged EXTRACTED (read from source) or INFERRED (derived by resolution).

Both claim to do retrieval without vectors; they differ in subject and output: one answers questions over long documents, the other produces a queryable graph of a codebase. The table compares item by item.

1. Open-source traction: VectifyAI/PageIndex has ~39K stars (verified 2026-10-13), created 2025-04-01; Graphify-Labs/graphify has ~126K stars (verified 2026-10-13), created 2026-04-03.
2. License: PageIndex is MIT; Graphify's repo is labeled Apache-2.0, with an additional LICENSE-MIT file at the root.
3. Retrieval mechanism: PageIndex: documents → tree index → LLM reasoning picks nodes; the official README says a basic model suffices for indexing and the chat model determines retrieval quality. Graphify: tree-sitter parses ~40 languages (vendor README), Leiden for community detection, docs and images via semantic extraction.
4. Benchmarks (vendor-reported): PageIndex claims 98.7% accuracy on FinanceBench financial-document QA (vendor-reported; eval code public in the Mafin 2.5 repo); Graphify's BENCHMARKS.md claims 0.497 recall@10 on LOCOMO and 76% QA accuracy on LongMemEval-S, run on the same harness against mem0 and supermemory (vendor-run, blind-validated kappa 0.81, self-reported).
5. Agent integration: PageIndex offers a local SDK (pip install pageindex, index locally with your own model key); a separate pageindex-mcp repo exists (MIT, HTTP transport), but the README's Local/Cloud table lists MCP server under Cloud features. Graphify offers a /graphify skill and an MCP server with tools including query_graph, get_node, get_neighbors, shortest_path, list_prs and get_pr_impact; the PyPI package is graphifyy (double y).
6. Cost: PageIndex says local tree indexing costs about $0.001/page, ~$1 for a 1,000-page document; Graphify's open-source engine is free with zero LLM cost for code parsing, while semantic extraction over docs spends the user's own model tokens (all vendor claims).