Small sized agents from Dria trained on interacting with an obsidian-like memory system using python tools. Trained on Qwen3-4B-Thinking-2507.