Persistent Graphs Eliminate Repo Rescans

AI coding agents like Claude, Cursor, or Codex waste tokens rescanning your entire repo for architecture, dependencies, and APIs on every query or context switch. Graphify solves this by generating a dynamic graph that tracks code structure, resolved bugs, and changes—serving as long-term memory injected into agent prompts. Result: Agents reference the graph instead of full scans, drastically cutting token use across sessions.

Generate the graph in your project root with /graphify . (or $graphify . in Codex). Link it to your agent via graphify <agent> install (e.g., Claude, Cursor). The graph auto-updates on code changes. Query it directly: /graphify query "what connects auth to the database?" or /graphify explain "RateLimiter". Extend with external knowledge: /graphify add https://arxiv.org/abs/1706.03762 (fetches and adds papers) or /graphify add <youtube-url> (transcribes videos). Export to Obsidian: /graphify ./raw --obsidian.

Caveman Skill Enforces Minimalist Outputs

Pair Graphify with Caveman, a skill that forces agents to respond in ultra-terse 'caveman' style—stripping unnecessary words for up to 75% token savings on every output. Applies automatically to chats after setup. Tune intensity: /caveman lite for mild brevity, full for aggressive, or ultra for extreme minimalism.

This combo targets the dual token drains: input context bloat from repo scans and verbose outputs. Agents stay efficient without losing core functionality, ideal for multi-agent coding workflows.

Frictionless Setup for Any Platform

Install Graphify: pip install graphifyy then graphify install (Linux/Mac), --platform windows for Windows, --platform codex for Codex, or graphify cursor install for Cursor. Full docs: https://github.com/safishamsi/graphify.

Caveman: npx skills add https://github.com/juliusbrussee/caveman --skill caveman. No Python needed. Works across supported agents for immediate token wins in daily coding.