Speak at Bengaluru Systems meet-up

Format: 20-minute talk with a demo.

An AI agent mostly just sits there waiting on the model. If you want hundreds of them on one machine, what counts is the memory each one holds while it waits. Rui is my harness in Zig: one process, one event loop, one HTTP/2 connection, with every agent stored as rows in SQLite. A hundred waiting agents fit in 7.4 MB. I’ll cover where other harnesses spend their memory, what’s left once you remove that (mostly libcurl bookkeeping, and one buffer flag took it from about 22 KB to about 6 KB per agent on Linux), how I measured it, and the trade-offs.

Links: https://github.com/DivyanshGolyan/rui and https://divyanshgolyan.com/posts/hundred-agents/