AI Nexus
AI coding assistant rule manager for Claude Code, Cursor, and Codex. Load only the rules you need.
Install
npx ai-nexus installclaude mcp add ai-nexus -- npx ai-nexus install{
"mcpServers": {
"ai-nexus": {
"command": "npx",
"args": [
"ai-nexus",
"install"
]
}
}
}Install command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →한국어 | English
ai-nexus
Only 2-3 relevant rules and skills load per prompt. The rest are hidden from Claude completely. Your rules and skills also work in Cursor and Codex — no extra setup.
Homepage | Documentation | Roadmap
npx ai-nexus install
The Problem
Rules with alwaysApply: true load on every prompt — the more you have, the more tokens you waste. With alwaysApply: false, Claude reads every rule's description to decide what to load — all your rule descriptions sit in Claude's context while it picks 2-3. You're paying Claude to filter rules instead of writing code. Skills and commands only load when invoked, but that means you have to remember what to call and when.
So most people only install a handful of rules. The more you have, the harder it is to know what's there — useful rules end up never loading or never getting installed in the first place.
A recent study from ETH Zurich (12 repos, 5,694 PRs) confirms this: loading all rules at once hurts performance by ~3% and increases cost by 20%+. The takeaway: only relevant context should be loaded per prompt.
On top of that, every AI tool has its own format — .claude/rules/*.md, .cursor/rules/*.mdc, .codex/AGENTS.md. If you use more than one tool, you end up maintaining the same rules and skills in multiple places, and they inevitably drift apart.
The Solution
ai-nexus filters rules before Claude sees them. A semantic router hook runs on each prompt, picks 2-3 relevant files, and physically parks the rest in rules-inactive/. Claude only sees what it needs — it doesn't even know the rest exist. Filtering is done by keyword matching (free) or GPT-4o-mini (~$0.50/month), not by Claude.
Nothing is deleted — just parked in rules-inactive/ and reactivated when the next prompt needs it. Write once, deploy across all your tools:
Write once:
config/rules/commit.md
config/skills/react.md
Deploy everywhere:
✓ Claude Code → .claude/rules/ (with semantic routing)
✓ Cursor → .cursor/rules/*.mdc (auto-converted)
✓ Codex → .codex/AGENTS.md (aggregated)
One source of truth. Every tool in sync.
Only 2-3 relevant rules and skills loaded per prompt.
Why ai-nexus?
Similar plugs
CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies
Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1
Documented system prompts from Anthropic - Claude Fable 5.1, Opus 5.5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-6-Astra, Codex. Google - Gemini 3.8 Flash, 3.1 Pro,…
Wrap Antigravity, ChatGPT Codex, Claude Code, Grok Build, Muse Code, Devin as an OpenAI/Gemini/Claude/Codex compatible API service, allowing you to enjoy the free Gemini…