Breaking

Open-source Graphify claims up to 71x token cuts for Claude Code

July 4, 2026 at 00:01 EDT

Software Dev & Coding
Open Source
AI Agents

Graphify, an open-source tool for AI coding assistants, is drawing attention as a way to cut token consumption when tackling large projects in tools like Claude Code. It turns a project's structure into a knowledge graph in advance, removing the need to re-read files at every session.

Open Source · Graphify

One map, not a thousand re-reads: cutting AI coding-agent token bills

Graphify builds a queryable knowledge graph of a whole codebase once, so later sessions reference the map instead of re-reading raw files — trimming the "orientation tokens" agents burn to understand a project by up to ~72×.

~72×

Peak orientation-token reduction (~71.5–72×)

~70%

Measured cut on the Formbricks project

500+

Files where the graph pays off most

Orientation tokens per task — same job, two approaches

Columns to scale (bookzero.ai example, ~27× lighter). Taller = more tokens burned.

27×

Re-read raw files
every new session

1×

Query the
knowledge graph

How the map replaces re-reading

Parse whole codebase
Tree-sitter reads every file once

→

Build knowledge graph
Real call graphs & import edges

→

Sessions query the map
No raw-file re-reads

Why builders like it

Runs locally — code never sent externally
Uses real dependencies, not vector search
MIT-licensed; incremental caching & Mermaid export
Works with Claude Code, Codex, Cursor, Gemini CLI, Aider

Open questions

Graph must be refreshed as code changes
Initial setup adds upfront effort
Package-name typo (graphifyy)
Reduction rates are self-reported — independent verification wanted

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.