Building My Own Plugin and Skill Marketplace: The Lemon AI Hub

The explosion of AI tools brought an unexpected problem: fragmentation. We have a skill for each task, a plugin for each platform, and our personal artificial intelligence ends up scattered across dozens of repositories and different contexts.

That's why I created Lemon AI Hub.

In this article, I'll tell you the "why" behind the project, how it was structured, and why you should consider having your own modular intelligence hub.

The Motivator: No More Reinventing the Wheel

The primary motivator was frustration with rework. Every time I started a new project, I found myself rewriting system prompts, reconfiguring API connections, or recreating skills I had already developed for other contexts.

I needed a place where:

1. My skills were portable: Working in VS Code (via Cursor), in the Terminal (via Gemini CLI), or in customized Dashboards.
2. Knowledge was cumulative: If I teach my AI how to review code following Karpathy's style, that skill should be available to all my agents.
3. The interface was brutalist and fast: No distractions, focused on performance and extensibility.

Architecture: Plugins as First-Class Citizens

Lemon AI Hub isn't just a collection of scripts. It was designed as a modular core.

The heart of the project is lemon-ai-hub. It works through a system of "Skills" and "Agents" that communicate via standardized protocols (like MCP - Model Context Protocol).

Example of a Modular Skill

A skill in the hub isn't just code; it's a manifesto of intent. Here's how we structured portability:

# Skill: Karpathy Pedagogy Reviewer
**Intent**: Reviews code with a focus on didactics.
**Trigger**: "review this code as Karpathy would"
**Workflow**: 
1. Find minimal examples.
2. Identify undefined jargon.
3. Suggest analogies (intuition pumps).

How the Marketplace Works in Practice

By centralizing everything on GitHub, I can "install" new capabilities into my local workflow with a simple git pull or activate_skill.

• Agents: Specialized personas (e.g., Frontend Designer, Architecture Reviewer).
• Skills: Atomic capabilities (e.g., Generate LinkedIn carousel, Optimize article SEO).
• Hooks: Automatic triggers that run on commits or state changes.

Spotlight: The Token-Saver Plugin 🛡️

One of the crown jewels of Lemon AI Hub is the token-saver plugin. It was designed for those who use LLMs at scale and want to prevent API costs from spiraling out of control.

token-saver configures compression and efficiency tools globally or per project:

• rtk (Rust Token Killer): The proxy that filters noise and optimizes calls.
• caveman: Ultra-compressed communication (up to 75% savings in prose).
• graphify: Codebase navigation via knowledge graph (cheaper than reading full files).
• context-mode: Intelligent context window management.

Real Results (rtk gain)

It's not just theory. The rtk gain command allows you to audit the exact savings. In one of my real-world use cases with structured search:

rtk grep: Savings of 358.7M tokens.

Yes, hundreds of millions of tokens saved through intelligent filtering and optimized routing. This transforms financially unviable projects into extremely affordable automations.

Why a Personal Marketplace?

Having a personal marketplace means you own your own AI "app store." You don't just rely on OpenAI or Anthropic to define the available tools. You create the tools that solve your pain.

If you want to see how this works in practice or start your own hub, check out the official repository:

🔗 Main Repository: github.com/Andersonlimahw/lemon-ai-hub

The future of productivity isn't about having one AI that does everything, but an ecosystem of specialized agents that you orchestrate yourself.

This post was generated and reviewed using Lemon AI Hub's own tools.