Hi, I am one of the authors of the linked article. For anyone interested in the implementation of the framework, here is the GitHub repository: https://github.com/matstech/dmf
That's an incredibly thorough analysis. And if we wanted to be even more picky: why on earth would we want to form so many opinions about the world using a single LLM model?
The idea for this library came while I was working on my graduate thesis. I found existing AI agent frameworks to be too complex and heavy for my needs.
I wanted more control over the prompts and tool calls, and less boilerplate code just to get started.
So, I took a DIY approach and built Linden: a lightweight library that provides the essential tools to connect LLMs to your code,
without forcing you into a rigid system.
A few key ideas behind it:
- Minimal Codebase: It has ~500 core lines of code. The goal is for you to be able to read and understand the entire library easily.
- No Vendor Lock-in: It supports OpenAI, Anthropic, Ollama, and Groq out of the box, making it easy to switch between providers.
- Full Control: The prompts are clean and work exactly as you'd expect. Tool invocation is predictable and reliable, with no hidden magic.
It's still early days and not production-ready, but I'd love to get your honest feedback and thoughts.