I'd love some feedback on a personal project. I've been building a lot of claw-like agent projects recently (memory, tools as the base). I abstracted out all the common things that I do in to an API that works similarly to Anthropic's Agent SDK but built on the back of Vercel's pluggable AI-sdk so you can use any model in the agent harness you want. https://github.com/PaulKinlan/agent-do Does this work for you? What would you change?