How I Built an AI Agent Toolkit That Sells While I Sleep

A few months ago, I found myself building the same AI agent scaffolding over and over. Authentication, memory management, tool calling, deployment pipelines—every new project started with copy-pasting boilerplate from the last one.

Sound familiar?

I decided to solve this once and for all. The result is a production-ready AI Agent Toolkit that I've now packaged and put up for sale. Here's the story of how it came together, what it includes, and why I think it might save you weeks of dev time.

The Problem: Reinventing the Wheel

Every AI agent project I've worked on follows the same pattern:

1. Set up project structure
2. Implement LLM integration (OpenAI, Anthropic, local models)
3. Build a tool system so the agent can actually do things
4. Add memory (because stateless agents are frustrating)
5. Handle errors, retries, and rate limits
6. Deploy and monitor

Steps 2–5 are pure undifferentiated heavy lifting. They're necessary, but they don't add unique value to your product.

I kept thinking: someone must have built this already. And yes, frameworks like LangChain exist. But they're often overkill, opinionated in ways that don't match my needs, or simply too abstract. I wanted something lean, modular, and easy to extend.

The Solution: A Toolkit, Not a Framework

The key insight was that I didn't need another framework. I needed a toolkit—a collection of well-designed, composable building blocks that I could mix and match.

Here's what's inside:

🔧 Core Agent Engine

A lightweight agent loop that supports multiple LLM providers (OpenAI, Anthropic, local Ollama models). Swap providers with a single config change.

🛠️ Tool System

A simple decorator-based tool registration system. Define a Python function, add @tool, and it's automatically available to the agent. No complex schemas to maintain by hand.

🧠 Memory Management

Pluggable memory backends: in-memory for testing, Redis for production, or even SQLite for simple persistence. The agent remembers context across sessions.

🚀 Deployment Ready

Docker configs, environment templates, and a GitHub Actions workflow for CI/CD. Deploy to any cloud provider in minutes.

📊 Observability

Built-in logging and tracing. See exactly what your agent is thinking and doing, which is crucial when things go sideways.

What Makes It Different

I've tried the big frameworks. They're powerful, but the learning curve is steep and the abstraction leaks are real. This toolkit takes a different approach:

• No magic: Every component is plain Python you can read and modify
• No vendor lock-in: Switch LLM providers without rewriting your agent logic
• No bloat: Only what you need, nothing you don't
• Production-tested: This is the code I use in my own projects

Who It's For

This toolkit is perfect if you:

• Are building AI agents and tired of starting from scratch
• Want to understand how agents work under the hood
• Need something more flexible than off-the-shelf solutions
• Value code you can read, modify, and own completely

The Business Side

Here's the part I'm most excited about: this toolkit is now a product. I listed it on LemonSqueezy for $9—less than the cost of a decent lunch.

👉 Get the AI Agent Toolkit on LemonSqueezy

Why charge at all? Two reasons:

1. It forces me to maintain quality documentation and keep the code clean
2. It validates whether this actually solves a problem for others (not just me)

What's Next

I'm actively using this in client projects and my own side projects. The roadmap includes:

• Multi-agent orchestration patterns
• Built-in web scraping and API toolkits
• More deployment templates (AWS, GCP, fly.io)
• A companion CLI for scaffolding new agent projects

Final Thoughts

Building products is hard. Building the infrastructure to support products shouldn't be. If you're working on AI agents and want a solid foundation to build on, check out the toolkit.

And if you do grab it, I'd love to hear what you build. The best part of creating something is seeing what others do with it.

---

P.S. The full source and more details are available at github.com/ulnit/agent-store