Introduce Kapybara: My Personal Agent

Let me introduce Kapybara, a general-purpose agent I built during the Lunar New Year holiday. I expect it to serve as a useful personal assistant. Compared with the agents people are already familiar with, it can work appropriately in group chats, distinguish between different chat windows, change its own avatar, schedule timed tasks, and plan tasks with DAG workflows. But beyond that, it does not yet have much that is distinctive at the product level.

I designed this agent more to test some of my ideas. In "How LLM Agents Become What They Look Like in 2026?", I explained my view of contemporary agents, and that soon made me want to build one myself.
After I started building it, I introduced a memory system based on my own intuition in "The 1st Memory Mode and the 2nd Memory Mode: Kapy's Memory System".
In "Requirements for AI-Native API Design", I explained how the work of building this agent differs from ordinary coding work.
In "Will Advances in Model Capability Crush All LLM Application Engineering?", I gave my answer to a question many people want to ask.
After finishing the current stage of Kapybara, I summarized the current design paradigm of general-purpose agents in "There Are No Agents, Only LLM".

All in all, I hope Kapybara and these essays can offer a bit of inspiration.