ShenSeanChen/waku-agent

Waku-agent is a personal AI agent designed to run locally, with the explicit goal of being readable — the author claims you can understand the full codebase in an afternoon. That is a meaningful constraint in a space where most agent frameworks have grown into labyrinths of abstractions. The stack covers the four things every production agent actually needs: a harness to run tools, a loop to chain steps, memory to persist context, and an eval layer to know whether it worked. Running locally means no API costs accumulating while you experiment, and no cloud dependency for dev iteration. The differentiator over Cindy or LangChain-style frameworks is the deliberate legibility — this is a learning artifact as much as a deployment target. Reservation: local-first also means you own the infra when you want to productionize it, and that gap is not bridged here. -> Best for: AI engineer or technical PM who wants to understand agent internals before committing to a framework.