🚀 Strategy Update: Why we are moving from EAT to LLM OS

[matiasmolinas]

Hi everyone,

First, thank you for the 445 stars on Evolving Agents Toolkit. It proved that we are all looking for the same thing: autonomous agents that can actually manage themselves.

We are moving development to a new repository: LLM OS

The Architectural Pivot

In EAT, we used MongoDB to store everything—including Agent Definitions. We found that this made "Evolution" (the ability for the AI to rewrite its own behavior) too rigid. It required complex CRUD operations just to change a system prompt.

The New "Hybrid" Approach

In LLM OS, we split the architecture:

1. The Mind (Markdown): Agents are now Markdown files. The LLM can read, write, and create new agents just by writing text.
2. The Body (Python): The kernel handles security (hooks), budget (token economy), and I/O.

What about the DB?

We removed it for v1 of LLM OS to reduce friction. However, we know file-based memory has limits. We are currently planning a specialized Advanced Memory Subsystem (likely Vector/Graph based) for LLM OS v2 to handle deep semantic relations.

We invite you to star the new repo and join the discussion on the new architecture!

👉 Link to LLM OS