Nice, I’ll take a closer look at that one too. I may be able to borrow some ideas from it. I haven’t used ouija for a big mesh yet. Mostly I use it as lightweight orchestration for sessions that are already running and already have useful context. The useful part is the ad-hoc handoff: I can be driving one session, then ask another session on my desktop, laptop, or another repo to help with a parallel thread. I also saw agentnoise yesterday from and had the same thought: there is probably overlap here so I'll try to check it during the weekend https://agentnoise.org/

With Claude Code programmatic usage moving into a separate credit bucket, I think terminal-native agent coordination gets more interesting. So I’m sharing ouija: https://github.com/dcadenas/ouija ouija is plumbing between already-running coding sessions: Claude Code, opencode, or both. No `claude -p`. No Agent SDK. Just normal terminal sessions in tmux. It shines for ad-hoc debugging: one session is deep in repo A, another already has context in repo B, and suddenly they need to talk. Same machine: tmux. Cross-machine: Nostr. The point is simple: let the sessions you already have talk to each other, even across machines.