jj

I'm building an agent harness library in C++. Nothing groundbreaking, just the usual agent interaction loop (with customizable tools) made available through a C++ API. 📋 Documentation: plywood.dev/docs/high-level/… 🐙 GitHub repo: github.com/preshing/plywood It's intended as an easy way to add AI features to other C++ projects. No need to pull in heavy JavaScript or Python frameworks; just use the C++ compiler that comes with your target platform. It's open source and works on Windows, Linux and macOS. A full-featured Markdown parser is included. It comes bundled with Plywood, my portable C++ runtime library. I think the main reason why C++ isn't more popular for agent development is because the standard library is too complicated. People also tend to worry about memory safety, but it isn't too hard to write memory-safe C++ if you follow simple rules. The advantages of using C++ are that it's supported everywhere, has a rich ecosystem of development tools, and of course performance. A sample application that streams conversations to a basic web UI is included. This sample is less than 30,000 lines of code (including the full Plywood runtime and Markdown parser) and the final executable is less than 1 MB in size. This project is a work in progress. So far it's compatible with four different inference providers, with more on the way.