When you run heavy multi-agent workflows, the bottlenecks start surfacing. There are plenty of ADEs and IDEs trying to fix them, but I don’t think agents need another interface. They need something like a DOS: an operating layer that keeps track of their tasks, their machines and what they’re allowed to do. Tasks are the processes, my devbox/teammates are other machines on the network (A2A/P2P), and the checks are the kernel.
The interface sits on top and can be anything. TUI/grok bot/instinctAI/Muse etc., or you can just have Claude build you a UI.
None of what kept breaking for me was a UI problem, so this repo is my attempt at fixing them in the layer underneath:
Repo: https://github.com/ELU-Labs/elu
npx -y u/elu-labs/elu
https://i.redd.it/1herxy3kv31i1.gif