CodebuffAI

The multi-agent architecture here is genuinely interesting — instead of one AI doing everything, CodebuffAI splits the work across specialized agents for picking files, planning changes, editing, and reviewing, which means the coordination layer is where the real magic (or mess) happens. It's open-source on GitHub, which is always a flex, and the approach is different enough from the single-agent assistants everyone else is shipping that it's worth watching — though I can't tell you if it's actually gained traction yet.